AI Will Finish What Neobanks Started — and Traditional Banks Won’t See Coming

The Pattern Every Disrupted Industry Follows

There’s a pattern to how established industries respond to technological breakthroughs. First, they observe from a safe distance. Then they hesitate, citing complexity or regulation. Eventually they adopt, but by the time they do, the clients worth keeping have already left.

Banking is living through exactly this moment and AI is the thing that will make it irreversible.

How Neobanks Moved the Goalposts

For years, the drift of tech-savvy corporate clients away from traditional banks was slow and easy to dismiss. Neobanks chipped away at the edges with better UX, faster onboarding, cleaner interfaces. But large banks could always point to stability, long-term relationships, and the inertia of enterprise procurement to hold their ground.

That argument is running out of road.

The businesses leaving first are doing so quietly. There’s no press release, no public breakup. Business account adoption has become a structural neobank trend, with the segment now accounting for roughly 67% of neobank revenue in 2025. This is driven precisely by companies that can’t afford the operational drag of legacy banking relationships. The writing on the wall: speed is non-negotiable for the operating layer now.

You cannot run a modern AI-driven business and tolerate a banking relationship where a wire transfer requires a manager to print a form, collect ten signatures, and manually key it back into a system. Consider what a single delayed wire transfer costs a company managing payroll across three currencies, or processing vendor payments tied to time-sensitive contracts. The mismatch goes beyond inconvenience. It snowballs across every transaction gradually until someone with a budget puts their foot down.

Why AI changes the question banks are being asked

Once you’ve rebuilt your business around AI, you see every vendor differently. You ask: why is this still manual? Why does this take days? Your bank is no exception. For most traditional banks, there’s no good answer.

The use case is concrete. A company running accounts payable through an AI agent needs infrastructure that can read an incoming invoice, determine the correct currency, trigger stakeholder approvals through an integrated workflow, and release the payment without a human intermediary at each handoff. This isn’t speculative. Finance teams are building exactly these workflows today, and every manual step their bank reintroduces at the end of the chain is a point of failure they’d rather eliminate.

A 2024 Accenture analysis projected that AI automation could reduce financial operations costs by up to 25% in treasury and payments. By late 2025, McKinsey’s Global Banking Annual Review put the number at 20% or more in net operational cost reductions from agentic AI alone, while also warning that these gains would largely be competed away rather than retained. A separate PwC analysis found that banks fully embracing AI could see up to a 15 percentage-point improvement in their efficiency ratio, with one institution reporting a 40% reduction in commercial client verification costs.

For businesses already achieving that kind of efficiency internally, a banking partner that reintroduces manual steps at the last mile is simply a liability at this point.

The Architectural Incompatibility Problem

Rather than just picking a bank, startups and technology companies assemble an operating ecosystem. Every tool in that ecosystem is expected to integrate, respond to new technology as it appears, and grow their operational efficiency over time. A bank that can’t provide a real-time balance (and remarkably, many of the world’s largest institutions still can’t) is architecturally incompatible with modern business infrastructure.

Why is this still the case? According to a 2024 report from 10x Banking, 55% of banks identify legacy system limitations as their single biggest barrier to achieving business goals, with more than half citing data silos and production bottlenecks as the reason they can’t scale. COBOL, the programming language developed in 1959, still powers more than 40% of core banking systems globally. 45 of the top 50 banks worldwide continue to run mainframes as their mission-critical infrastructure. The original developers have mostly retired, and the institutions running this code often lack the internal expertise to fully understand what it does.

It’s not that traditional banks don’t want to modernize, but that incremental patching of a 60-year-old core cannot produce the API-first, event-driven infrastructure that AI-native businesses need as their banking layer. You can’t just retrofit a batch-settlement system to behave like real-time infrastructure since these architectural constraints are fundamental.

Traditional banks learned to offer card payments. Then mobile apps. Then, eventually, some form of API access. Each time, they treated the new capability as a destination rather than a direction by implementing it, declaring victory, and then falling behind the next curve.

The institutions that respond by bolting an AI chatbot onto a legacy core will find themselves in the same position they were in when neobanks first appeared i.e, they will watch clients leave without understanding why.

Who Leaves Next — and When

The businesses that moved first (AI-native startups, crypto-adjacent fintechs, tech operators) have largely made their decisions. The second wave will be the mid-sized businesses and larger corporations who have already felt AI reshape their own industries. Whether through internal automation that altered their cost structures, or through competitive pressure that changed their markets entirely.

The loyalty shift is already measurable. McKinsey’s 2025 Global Banking Annual Review noted that in the United States, only 4% of new checking account openings now come from existing bank customers — down from 25% in 2018. That is not a blip but a structural unraveling of the inertia traditional banks have long relied on to hold their client base.

The same report projects that banks failing to adapt could see their global profit pools decline by $170 billion, roughly 9%, over the next decade. More strikingly, the threat McKinsey identifies doesn’t come only from neobanks or fintechs. It comes from customers themselves using AI agents to optimize their own finances: moving deposits to better rates, managing credit utilization, routing payments through better infrastructure. A client that builds an AI-native treasury function internally doesn’t need their bank to do it for them. Matter of fact, they need their bank to stay out of the way.

The Dividing Line

The coming divide in banking is between banks that were built for this moment and banks that are trying to adapt to it. Not large and small institutions. Not incumbents and challengers.

Built-for-it means the API layer is the product instead of a bolt-on. Real-time infrastructure is the current operational reality. Compliance workflows, FX execution, and approval logic are all programmable by the client’s own systems now instead of being routed through a relationship manager’s inbox.

According to The Financial Brand’s 2025 Retail Banking Trends report, only 25% of banks have prioritized modernization of their back-office infrastructure, even as more than half list digital experience as a strategic priority. That gap, between stated intention and actual architectural investment, is exactly where the next wave of client exits will originate.

Neobanks proved a better experience was possible. AI will prove that the human-in-the-loop model of banking is no longer viable for the businesses moving fastest. For the banks that waited too long, the window will close all at once, in the way these things always do. Slowly, then suddenly.

The more interesting question for me now is whether there is a traditional institution with the organizational will to move before it happens, or is the gap between strategy decks and actual infrastructure too wide to close in time?