Why Financial Institutions Are Still Reading the AI Manual

It is a truth universally acknowledged that if you give a financial institution a new piece of technology, it will first attempt to form a committee to decide which font should be used for the warning labels. This is not a criticism, mind you; it is simply the natural order of things, much like the way a cat will always choose the most expensive piece of furniture to sharpen its claws upon. However, in the current climate of fintech innovation, the cats have not only sharpened their claws but have also learned to operate the vacuum cleaner, while the furniture is still trying to remember where it put its spectacles.

Recent reports from the front lines of the 'AI Adoption Race' suggest that the gap between what is technically possible and what is institutionally manageable has reached the proportions of a small, but very deep, canyon. On one side, we have the fintech startups, entities that seem to be composed entirely of caffeine and algorithms, deploying 'agentic AI' that can process a mortgage application in the time it takes a human to blink. On the other side, we have the traditional institutions, which are currently engaged in a very serious debate about whether an AI agent counts as a 'person' for the purposes of the annual Christmas card list.

The problem, as it turns out, is not that the technology is too complex. It is that the technology is too fast. A traditional bank is a marvel of engineering, designed to move with the majestic, unhurried grace of a tectonic plate. It is built on layers of legacy systems, some of which are so old they are rumored to be powered by steam and the occasional sacrifice of a junior clerk. To introduce a thinking, self-correcting algorithm into this environment is a bit like inviting a hummingbird to a meeting of the Royal Society of Tortoises. The hummingbird is finished with the agenda before the Chairman has finished clearing his throat.

I once observed a particularly ambitious bank attempt to integrate a basic chatbot into its customer service department. The chatbot, eager to please, began resolving queries with such efficiency that the human staff became suspicious. They eventually decided to 'throttle' the AI, forcing it to wait exactly three minutes before responding to any message, so as not to startle the customers. There is something profoundly human about the desire to make a machine appear more incompetent just to make ourselves feel more comfortable. It is the digital equivalent of putting a hat on a robot so it looks less like it's planning to take over the world.

The 'Institutional Readiness' gap is often framed as a failure of regulation, but that is a rather uncharitable view. Regulators are, by their very nature, the people who check the brakes on the train. The issue is that the train has currently been replaced by a teleportation device, and the regulators are still looking for the wheels. In Nigeria, for instance, the Central Bank has recently mandated that fintechs use AI for anti-money laundering compliance, giving them a generous two years to comply. Two years in the world of AI is roughly equivalent to four geological epochs. By the time the deadline arrives, the AI will likely have retired to a small cottage in the Cotswolds to write its memoirs.

There is also the matter of the 'Playbook.' Companies are now releasing comprehensive guides on how to deploy agentic AI in financial services, as if one could simply follow a set of IKEA instructions to build a sentient ledger. 'Step 1: Insert the Large Language Model into the existing database. Step 2: Ensure the ethical guardrails are tightened using the provided Allen key.' In reality, the database is usually a tangled web of COBOL and hope, and the Allen key was lost in 1984.

The result is a sort of polite, high-stakes comedy of errors. We see fintechs launching products that promise to 'democratize finance' through automated intelligence, while the institutions that are supposed to provide the underlying infrastructure are still trying to figure out how to stop their printers from jamming. It is a race where one participant is wearing rocket boots and the other is still trying to find a matching pair of socks. The rocket boots are, of course, very impressive, but they do tend to set fire to the carpet.

Perhaps the most whimsical aspect of this entire affair is the way we talk about 'readiness.' We speak as if there is a specific moment when a bank will wake up and say, 'Yes, I am now ready for the singularity. Please pass the marmalade.' But readiness is not a destination; it is a state of perpetual catching up. The manual is being written while the machine is already running, and several of the pages are currently on fire.

In the end, we may find that the only way for institutions to truly become 'ready' is to stop reading the manual altogether and simply start pressing buttons. It is a terrifying prospect for anyone who values the stability of the global economy, but it is also undeniably exciting. After all, there is nothing quite like the thrill of watching a multi-billion dollar organization try to explain to a regulator why its AI accidentally bought a controlling stake in a company that makes novelty hats for pigeons. It is the sort of progress that keeps life interesting, even if it does make the quarterly reports a bit more surreal.

As we move forward, the gap will likely continue to widen. The fintechs will continue to innovate at the speed of light, and the institutions will continue to process the implications at the speed of a particularly thoughtful snail. And perhaps that is for the best. A world where everything moves at the speed of light would be very bright, but also very exhausting. There is much to be said for the snail, provided it doesn't get stepped on by someone in rocket boots.