The bank of one: Provocation or prophecy?

Finance, after all, is an industry built entirely from information. It is nothing but information, which places financial services in a not entirely unique group of industries that are fully visible and accessible to AI. If everything can be done remotely and via computer, it can be done by AI. Or can it?

Even as AI seeps into every corner of financial services – from instant credit decisions to personalised financial advice – the notion of a fully automated "bank of one" still feels uncomfortable, perhaps even reckless. It is only natural to object. But what, precisely, stands in the way?

A question of perspective

The first protest is often straightforward: "You can't just ask ChatGPT to run the bank!" Indeed, nobody would trust today's AI to manage an entire institution autonomously. But zoom in on any banking process – processing loan applications, drafting compliance reports, or handling customer inquiries – and the narrative changes.

Every task, examined closely, consists of simpler tasks, each of which current AI can handle confidently, especially with a bit of “scaffolding” in the form of traditional software support to guide the process. Goldman Sachs has already introduced AI assistants to thousands of employees to streamline research and compliance tasks – with astonishing success. Each model upgrade shortens the gap between partial automation and complete autonomy.

This is not to say that banks should immediately hand over the keys to an AI. Instead, it raises a challenging question: If AI can handle nearly every individual process, what is left that genuinely requires human intervention?

But surely, banking is not that simple?

The next objection: banking involves more than routine tasks. True strategic insight, after all, demands human creativity, experience, and judgement. It is easy to accept that much. Yet, how extensive is the strategic portion of banking, really?

Maybe some especially high-stakes customer interaction or the most delicate interactions with regulatory authorities and biggest strategic calls in investment banking are still beyond the scope of AI, requiring human expertise and experience at a level beyond current AI models. But the models’ capabilities are growing fast. The original 2024 version of ChatGPT scored as well as a typical grad student on many exams. The current 2025 version scores deep into the upper 90^th percentile and shows no sign of slowing down.

In many day-to-day operations of financial services, deep expertise is not always necessary. For efficiency and consistency, these processes must be scalable, standardised, and repeatable.

Finance: A human business… Or is it?

Another line of defence might sound familiar: people want human relationships. Financial institutions have long marketed themselves as relationship-driven businesses, but do customers truly crave that personal touch?

Emerging fintechs and neo-banks, such as Switzerland’s own neon, challenge that assumption every day. They win customers not with handshakes or famously generous bank affiliate opening hours, but with instant service, superior convenience, and lower fees. Moreover, customer satisfaction scores for AI-driven services are rising quickly, boosted by round-the-clock availability and perfect recall of customer preferences. AI is not merely approaching human emotional intelligence – it may even be starting to surpass it.

People already spend hours each day interacting with generative AI purely for enjoyment, turning to AI platforms for companionship or stress relief. Why, then, would they object to handling routine financial tasks in the same way?

Regulation: Protector or prison?

Perhaps the most potent shield against automation is regulatory. It is entirely legitimate to suggest that regulators would recoil from a highly automated "bank of one," and understandably so. Stability, accountability, and explainability are non-negotiables for financial institutions.

Yet regulation is not a static barrier; it is a negotiation. Regulators worldwide, from Switzerland's FINMA to the EU’s AI Act, are issuing guidelines not to forbid AI, but to manage it safely. The constraints they impose – such as audit trails, transparency, and accountability – are not designed to prevent automation, but rather to shape how it is done responsibly.

I will concede that I personally see this as the most likely barrier. Change in the financial services industry is slow – and for good reason. Many of these institutions have worked well for hundreds of years, and mistakes can have real consequences for people's well-being. A high price for change is justified simply in the name of care and safety.

But even this is a brittle defence. Regulators are changing. And the industry is also facing pressure from elsewhere. Could neo-banks – unburdened by legacy systems and built with AI compliance at their core – reshape both regulatory expectations and customer demands?

The real provocation

The "bank of one" is not an imminent reality – it is, however, a useful provocation that challenges the comfortable assumptions underpinning the finance industry's sense of security. If finance is fundamentally about information and judgement, and if AI excels precisely in that realm, how long can established banks resist profound automation?

Banking leaders and regulators alike must confront an uncomfortable truth: protecting jobs through regulation or nostalgia for the human touch is unlikely to be a sustainable strategy. The real defence against AI-driven disruption lies in demonstrating genuinely unique human value – not merely asserting that it exists. Or perhaps AI will reshape the industry just as digitalisation and the internet once did. If the thesis presented here holds true, a new neo-bank would have an organisational chart almost unrecognisable to a traditional bank – and it would be a force to be reckoned with, both in terms of service and price.

The "bank of one" might never literally open its doors – but banks with far fewer employees, dramatically lower operating costs, and vastly superior customer experiences might. If this scenario feels unrealistic, uncomfortable – or even threatening – then perhaps that is exactly the conversation the industry needs to have now, before it becomes unavoidable.