Why financial institutions can’t deliver on AI, compliance, or strategy without rethinking their architecture.
For years, banks have been told to transform. Modernize tech stacks. Rethink customer experiences. Move faster, integrate smarter, embed everywhere. Now add one more: become AI-native.
But despite ambitious roadmaps and record investment, many financial institutions remain stuck. Not for lack of vision, capability or funding, but because the systems they are building on were never designed for change.
Engineers are drowning in integration backlogs. Product launches stall. Leadership is caught between patching outdated systems and committing to multi-year transformation programs that overpromise and underdeliver. And the AI pilots that were supposed to change everything in 2025 are, for most institutions, still pilots.
The blocker is not capability. It is architecture. Banks have answered fragmentation three ways. Keep patching, and the silo count grows. Replace the core, and the first win is years out. Wrap the estate in cloud and APIs, and the surface changes while what sits beneath it does not. Until a fourth move takes hold, transformation will keep falling short.
Legacy systems, even those wrapped in cloud, were built for a different era, one that prioritized stability over agility and control over flexibility. Products launched annually, customer expectations were predictable, and quarterly release cycles were enough to stay relevant.
That era is over, but its bills keep arriving. Capgemini’s 2026 research into corporate and investment banks found that 43% of IT budgets go to running and maintaining legacy systems, against 29% directed to transformative technologies.1 Every dollar spent keeping the old estate alive is a dollar not spent on the products customers are already asking for.
There is a quieter cost, and it compounds. In a conventional estate, every product ships its own back office. The lending system has its own audit trail, approval chain, task queue, policy configuration and permissions, and so does the payments hub, the onboarding tool and the treasury workstation. A bank running twelve products runs twelve audit trails and twelve approval frameworks, and its compliance surface is the sum of them. Each new product makes the next one harder to launch and harder to govern.
Today’s market does not move in predictable cycles. Banks are expected to launch products at speed, respond to market shifts overnight, and connect across fintechs, platforms and partner ecosystems. Real-time finance demands real-time financial infrastructure, and that level of adaptability cannot run on rigid, siloed systems.
Today’s most profitable new financial providers did not win by offering radically different products. They won by offering them at the right time, and by iterating fast enough to stay relevant.
In a market where customer needs shift quickly and margins are under pressure, time-to-market has become a proxy for competitiveness. Nearly three in five corporate banking clients now expect real-time responsiveness from their banks, yet fewer than one in four say those expectations are being met.1 The providers able to respond to demand in weeks, not months, earn that loyalty repeatedly.
Incumbents know it. In KPMG’s 2026 Banking Technology Survey, 71% of US banking leaders agreed their organizations need to invest in modernizing platforms to bring new or enhanced products to market, up from 46% a year earlier.2 That is a shift in diagnosis as much as ambition. Leadership has started naming the platform, rather than the roadmap, as the constraint.
For larger institutions, the challenge is not a lack of ambition. It is the inertia baked into complex systems, where every new product requires orchestration across teams, platforms and approval layers that were never designed to move together. The widening gap between what strategy requires and what systems allow is where the next wave of differentiation will play out.
Ask anyone in finance what will define their business over the next five years, and the answer is still AI. The word in front of it is now “agentic”. The ambition is real: autonomous fraud detection, dynamic pricing, agents that onboard customers and process loans end to end. The infrastructure to support it is, for most, still stuck in the past.
The numbers tell the story. In Capgemini’s World Cloud Report in Financial Services 2026, 80% of financial services firms were at the ideation or pilot stage with AI agents, and 10% had deployed them at scale.3 Three-quarters of banks plan to put agents into customer service, and 61% of executives call cloud-based orchestration critical to their AI strategy.3 Among corporate and investment banks, 82% of executives said their innovation efforts were not producing improved revenue through new products, and 51% said they had not delivered the expected cost savings either.1 In the US, 80% of banking executives expect AI to significantly disrupt their business and operating models within three to five years.2
Expectation is not the problem. Look instead at who the 10% are, and at what they built first.
At Sibos 2026, BNY described a digital employee that repairs payment instructions failing straight-through processing because of beneficiary, reference-data, address or purpose errors. It now handles more than a tenth of the bank’s payment repairs globally, and a person still makes the release decision.6 BNP Paribas put an agent in charge of dispatching work across its securities trade processing, cut the workflow from ten steps to six, and had the agent doing most of the work within weeks, with every model interaction running through an observability layer carrying guardrails set by compliance, legal and HR.6 HSBC’s agentic trade-document checking is live in Hong Kong, the UAE and the UK, with people handling exceptions and higher-value transactions.6 Deutsche Bank reported onboarding a client and issuing a loan in a day instead of a month.6
Three things are true of every one of those deployments. The institution was large enough to build its own orchestration and control layer first. The agent acts through that layer, on shared customer and payment state, under one audit trail. And a typed human decision stands in front of the money.
That is the pattern the other 90% cannot build for themselves, one bank at a time, on an estate of siloed systems, dirty data and brittle integrations. An agent can only act on what it can see, and it can only move as fast as the systems it orchestrates. Until financial architecture handles event-driven processes with governance built in, AI will remain a pilot rather than a differentiator.
When this article was first published, DORA, the Digital Operational Resilience Act, had just come into force and ISO 20022 was still a deadline on the horizon. Both have now moved from preparation to proof.
In June 2026 the European Supervisory Authorities published their first DORA incident report. Financial entities in the EU reported 3,383 major ICT incidents in 2025, DORA’s first year, around a third of them with cross-border impact.4 Only 10% were cybersecurity-related. The main drivers were system failures and external events: the plumbing, and the third parties the plumbing depends on.4
The United Kingdom gave that finding a face in 2026. In March, a software defect introduced in an overnight update at Lloyds Banking Group let customers see other customers’ transactions, followed by goodwill payments and questions from Parliament’s Treasury Committee. In June, customers of Lloyds, Halifax and Bank of Scotland were locked out of their apps and online banking for several hours, unable to pay or transfer.7 Neither was an attack. Supervisors have responded by shifting from reviewing policies to demanding evidence: real-time proof of resilience, automated reporting, and demonstrable control over ICT risk, including the concentration risk that sits with third parties.
ISO 20022 tells a similar story. Swift retired cross-border MT payment instructions on 22 November 2025, ending the coexistence period for cross-border payments.5 Many institutions got there on translation layers rather than native adoption. Where a payment moves between ISO 20022 and MT mid-chain, structured information is truncated or lost, and that affects compliance screening, investigations and downstream processing.8 The four errors BNY’s repair agent spends its day on, beneficiary, reference data, address and purpose, are the kind of data a translation layer can truncate or drop.6,8 The structured-address requirement that follows, whenever Swift’s revised release schedule lands, means core upgrades, data cleansing, and reworked sanctions and AML screening.
These frameworks demand dynamic risk controls, faster auditability, and policy orchestration at scale. Yet most compliance processes remain hard-coded and manually maintained, which makes every update slow, costly and complex. Capgemini found that 96% of financial services executives cite regulatory and compliance burden as a roadblock to AI adoption, and 61% of corporate and investment banking executives say they are constrained by high compliance costs.3,1 Siloed data, overlapping risk controls and disconnected workflows turn compliance into a high-stakes stress test that patching and workarounds cannot fix.
The alternative is to embed policies, risk controls and workflows directly in infrastructure that is dynamic, orchestrated, and auditable by design. Then compliance stops being a bottleneck and becomes a strategic advantage. This is the direction leading financial institutions are moving in:
In the AI and Real-Time Era, execution will not be driven by standalone software or point solutions. It will be driven by financial infrastructure that adapts to business intent, customer behavior and regulatory shifts without slowing down.
Banks have tried three answers to fragmentation: keep patching, replace the core, or wrap the estate in cloud and APIs. The fourth move is different: deploy a substrate above what already runs. The estate stays. The fragmentation goes. Integrate once, and reuse compounds.
Adaptive Financial Infrastructure (AFI) names that move. The AFI Platform is XYB’s banking orchestrator: event-native infrastructure, built on Apache Kafka and Apache Iceberg, unifying systems, people and processes on one substrate. Here is how it answers the pressures above.
Systems integrate once. A core, a rail, a KYC provider or a fraud engine integrates once, to a canonical model, and every capability on the platform reuses that integration. Vendors sit behind adapter contracts, so the platform binds to no vendor’s roadmap, pricing or geography, and an outage in a third-party API is contained at the integration boundary instead of cascading into the event stream. That is the DORA finding on system failures and third parties answered in the architecture rather than in a policy document. And the platform itself is SOC 2 Type II attested and ISO/IEC 27001:2022 certified, so the third party a bank adds is one whose controls are independently examined.
Operations govern once. Workflow, rule, policy, task, approval and audit event are platform objects, not features each product rebuilds. Because every capability’s state changes are events on one stream, the audit trail is a property the substrate confers, by construction. Add a capability and it arrives governed on day one: same audit spine, same approval discipline, same operator inbox. The bank’s compliance surface stops growing with its product count.
Every action carries a verdict. Before any action that changes money or account state, one evaluation returns one typed verdict: allow, deny, hold, step up or refer. Every payment, on every rail, reports through one status model, so ISO 20022’s structured data is native to the platform rather than translated at its edge.
Put those three together and the agentic picture changes. An AI agent is a capability like any other. It arrives with the same audit spine and the same approval discipline, acts on the same shared state, and the human release that BNY, BNP Paribas and HSBC each built for themselves is already a platform object. Agents, policies and workflows share one orchestration layer, the same one DORA supervisors and ISO 20022 counterparties are now asking banks to prove they have.
The platform is open and built to work alongside existing systems, unifying what already runs. The adoption path is Deploy, Expand, Modernize: land the first workload in weeks instead of committing to a multi-year program, add capabilities, segments and regions on the substrate already in place, then move more of the stack across at the bank’s own pace. The platform orchestrates more than one million transactions per minute, and institutions see first production value in ten weeks.
Consider a bank running an established core. The core stays the system of record. The platform takes on payment origination and the file exchange with the bank’s ACH and wire processing, with approvals, audit and reconciliation handled once in the orchestration layer rather than rebuilt for each flow. The first workload goes live without a replacement program, and each capability added after it arrives already governed.
Derek Joyce, CEO at XYB, sums up the core challenge facing financial institutions today:
“XYB exists because banks can’t keep building tomorrow’s products on yesterday’s systems. Our mission is simple: create infrastructure that adapts, thinks, and evolves in real time, so financial institutions can stop patching and start progressing.”
For years, banking transformation has been framed as a choice between two unsatisfying paths: patch the legacy system or replace it entirely. The last eighteen months have made plain that neither goes far enough or fast enough. AI budgets rose, pilots multiplied, and 82% of corporate and investment banking executives still cannot point to the revenue.1
The institutions that will define the next decade of finance will not be the ones with the biggest budgets. They will be the ones with the clearest execution path for delivering adaptive products: infrastructure flexible enough to adapt, intelligent enough to respond, and open enough to scale alongside change.
The banks that have already scaled AI agents built that layer for themselves. Adaptive Financial Infrastructure is the same move made available to every other institution, above the systems it already runs. That is the shift AFI names and the AFI Platform makes possible.
Aug 17, 2026 · 14 min read
Aug 20, 2026 · 3 min read
Aug 17, 2026 · 10 min read
New to Adaptive Financial Infrastructure? Start with Why AFI. Ready to see it live? Book a demo.