Banks have spent years promising that AI will change everything. Now it is, and the truth is uncomfortable. Most banks are not ready for what is actually coming.
The divide splitting the banking world right now is not about which institution has the best AI tool. It is about who has built systems that can truly use AI the way it was designed to work. That gap is growing fast, and the stakes could not be higher.
The Difference Between Using AI and Being AI-Native
Not every bank using AI is an AI-native bank. That distinction matters far more than most executives want to admit.
Using AI to speed up a marketing offer or flag a suspicious transaction is useful. It is not a transformation. A truly AI-native bank is one where AI sits at the core of every financial decision, every product, and every customer interaction, not as a tool bolted on from outside, but as the foundation the entire system is designed around.
A Q1 2026 survey by Accenture found that 91% of banking executives consider AI a strategic priority. But only 23% have moved beyond pilot programs into actual production deployment. That gap between ambition and action is where most banks are quietly losing ground right now.
IBM data from late 2024 showed that only 8% of banks were developing generative AI in a truly strategic, enterprise-wide way, with the remaining 78% still operating in what researchers called “tactical mode.”
McKinsey’s latest report, “The AI-powered bank: Rewiring for excellence in customer care,” is even more direct. It states that the gap between leaders and laggards “is no longer defined by the tools they buy, but by the operating models they build.” That single line says everything about where banking stands today.
Why Legacy Systems Are the Real Problem
The honest answer to why most banks are stuck is simpler than it sounds. Their systems were never built for this.
43% of financial institutions still run on legacy systems built more than 20 years ago, and maintaining those aging systems now consumes up to 70% of annual IT budgets.
That leaves almost no room for real transformation. Here is what those old systems fundamentally cannot do in an AI-first world:
- Process continuous, real-time data streams that AI models need to function
- Share data freely across departments without friction and delays
- Support autonomous AI decision-making at scale
- Adapt quickly to fast-changing regulatory requirements
- Connect to modern cloud-native AI platforms without costly workarounds
Visa’s global head of issuing solutions, Kathleen Pierce-Gilmore, called trapped data inside aging banking infrastructure “data prisons,” places where critical information sits locked inside systems that cannot move it in real time. AI depends on timely, governed access to data. Legacy batch-based environments simply cannot deliver that.
McKinsey put it bluntly in its recent report: “Most banks are not failing because their technology is weak; they are failing because they are layering AI onto legacy operating models and automating broken processes rather than redesigning them.” Banks have poured money into voice bots, agent copilots, and real-time analytics with promises of cutting costs by 30 to 45%, but those gains have largely remained unrealised.
Microsoft’s corporate vice president of worldwide financial services, Bill Borden, warned in April 2026 that legacy systems will face mounting pressure as AI-driven, machine-executed transactions scale up across the industry. The core challenge, he said, is no longer what AI systems can do, but whether they can be trusted, audited, and controlled inside highly regulated environments.
Real-World Banks Already Making the Move
Some institutions are not waiting. They are rebuilding right now, and the early results are hard to argue with.
JPMorgan Chase has allocated $20 billion of its total $105 billion 2026 technology budget to AI and related infrastructure. The bank has also given 200,000 employees access to its internal LLM Suite platform, with the bank now running more than 450 active AI use cases across origination, capital markets, and operations.
Bank of America’s AI assistant Erica has now crossed 2.5 billion client interactions, providing proactive insights for 20 million customers. Goldman Sachs has gone a step further, deploying autonomous AI coding agents across its developer workforce and treating them as what one executive described as “digital employees” rather than just software tools.
Oracle launched its dedicated agentic banking platform in February 2026, describing it not as a collection of point solutions, but as a foundational architecture for building “truly intelligent banks.”
Financial institutions that have fully redesigned their operations around AI are already reporting cost reductions of nearly 20%, according to McKinsey’s Global Banking Annual Review 2025. McKinsey also projects that AI could add up to $2 trillion in total annual value to global banking when you account for revenue gains, risk reduction, and new product creation.
For banks not ready for a full overhaul, IDC projects that 40% of global banks will run a “sidecar” strategy by end of 2026, operating a modern AI-ready core alongside existing infrastructure while migrating functions over time. By 2028, that number is expected to hit 70 to 80%.
Who Controls Decisions When AI Is Running the Show
This is the question the industry is not asking loudly enough, and it may be the most important one of all.
When AI moves from suggestion to execution, traditional governance models start to break down. Banks have always managed risk by identifying the exact moment a decision is made and ensuring a human is accountable for it. Continuous AI execution does not work that way.
The Bank for International Settlements has already flagged this directly, noting that as AI systems move closer to execution, risk and control can no longer be managed purely through oversight. They have to be designed into the system itself.
McKinsey made this even clearer: “True value is unlocked only when risk and compliance are treated as integrated design constraints rather than a final gate to be cleared.”
Microsoft is now developing tools that assign identities and permissions to AI agents and track every action they take. In regulated environments, firms must be able to show what controlled a system and whether it followed policy when a machine made the call without human input. That level of traceability is the baseline, not an optional extra.
The UK’s Financial Conduct Authority has taken the accountability approach even further. Under its Senior Managers and Certification Regime, if an AI model discriminates against a borrower, regulators do not just fine the institution. They look for the specific executive who signed off on that model. That kind of individual accountability is now directly shaping how boardrooms across the industry think about AI deployment.
The Cost of Standing Still
For banks still treating AI as an experiment, the clock is running faster than it looks.
McKinsey’s warning is clear: “AI implementation is not a chatbot initiative. It is a fundamental operating model transformation.” Banks that keep automating broken processes will only amplify their existing problems, not fix them.
| AI-Native Banks | Banks Layering AI on Legacy |
|---|---|
| Up to 20% operational cost reduction | Marginal or no savings realised |
| 25 to 40% fewer customer service calls | AI amplifies existing inefficiencies |
| Real-time autonomous decision execution | Batch-based delays persist |
| Compounding competitive advantage over time | Widening gap versus digital-first rivals |
PwC analysis found that institutions fully embracing AI could drive up to a 15-percentage-point improvement in their efficiency ratio. McKinsey adds that early movers stand to gain a 4-point return on tangible equity advantage over those who delay. Those are not small numbers in an industry where every basis point matters.
The pressure is not only coming from within traditional banking. Neobanks are scaling globally on lightweight, cloud-native infrastructure with none of the legacy weight slowing them down. Tech giants are quietly embedding payments, credit, and savings directly into everyday consumer apps. Nearly half of digital banking users say they would switch providers for a better digital experience, and a growing share already have.
The transformation of banking is no longer something that is coming. It is happening now, inside institutions that are willing to face the hard structural choices others keep putting off. The McKinsey report released just weeks ago made the stakes painfully clear: success will belong to banks that have the courage to dismantle the legacy structures holding AI back, not just the ones that spent the most on tools. Every month of delay has a measurable cost, and those costs compound. What do you think: can traditional banks actually pull off this kind of deep rebuild, or are the digital-first challengers already too far ahead? Share your thoughts in the comments below.
