Interviewed by Emmanuel Daniel
Stablecoins are raising pressure on bank payments and deposits, pushing ANZ and UOB to focus less on the form of digital money and more on client liquidity, settlement, asset distribution and interoperability.
ING, EastWest Bank and ANEXT Bank are taking different paths through core modernisation as modular systems, governed data, production controls and closer links between technology and the business take on greater importance in putting artificial intelligence into everyday banking.
Artificial intelligence (AI) is raising the stakes for core modernisation. Core modernisation increasingly requires banks to keep systems, data and controls adaptable over time. Banks are also under growing pressure to show that AI can create business value beyond pilots. That increases the importance of the underlying architecture because AI needs timely access to systems, data and workflows before it can act within banking processes.
That requirement links the approaches described by Daniele Tonella, chief technology officer and management board member at ING, Barani Sundaram, chief technology officer at EastWest Bank in the Philippines, and Jackson Oh, chief technology officer at ANEXT Bank in Singapore. They described architectures that allow individual capabilities to change, data to be accessed through controlled routes and safeguards to remain in place as AI use expands. ING has spent more than a decade breaking up a traditional core, EastWest is using application programming interfaces (APIs) to work around systems it can retain for now, while ANEXT began with a cloud-native platform. Their starting points differ. Each treats modernisation as an ongoing operating discipline.
ING draws on earlier architectural choices as AI use expands
ING’s experience shows how architectural changes made long before the current wave of generative AI can shape the introduction of new capabilities. Tonella described a programme that began more than 10 years ago with modular architecture, APIs and microservices, or independent software components that can be changed without rebuilding the whole system. “That led to the disappearance of what used to be called the core banking concept,” he observed. ING reduced the traditional monolith towards an account management system, with other functions organised as modules around it, and now draws on that structure as it expands AI use, with a few major cases already in production and a couple of hundred moving closer to production at the time of the discussion.
That architecture gives ING a controlled route for AI to access bank functions and data. Tonella noted that ING placed Model Context Protocol, a standard that allows AI systems to call external tools and services, over its existing API architecture, so calls can continue to pass through existing identity and authorisation controls. He also linked AI readiness to governed access to the bank’s data and contextual information, keeping data governance within the same architectural problem.
Tonella also cautioned that APIs require strict architectural governance. Without it, overlapping interfaces can create another layer of complexity and undermine the benefits of modularisation. The practical question is which capabilities and data can be accessed or changed independently while identity, authorisation and governance remain consistent. ING reduced the amount of functionality trapped inside the core. EastWest is pursuing similar flexibility while retaining systems that remain fit for purpose.
EastWest makes coexistence part of the modernisation strategy
EastWest approaches the issue as an incumbent bank with systems at different stages of their useful life. Sundaram rejected age as the defining test for legacy. “Legacy is anything that stops the bank from changing safely or innovating quickly,” he argued, adding that the constraint can sit in a process or organisational structure as well as a system. EastWest assesses platforms against three practical questions: whether they still protect customer trust, whether they allow the bank to move at the speed the market demands and whether they remain economical to operate. Systems that remain stable, secure and fit for purpose can stay, while APIs decouple them from new customer journeys and product development.
The bank has built more than 600 microservices around this approach and is continuing a multi-year core programme. Sundaram called the method “API-only” because the bank can work around systems that still have useful life while addressing the parts that genuinely constrain the business. Coexistence allows EastWest to separate the pace of product and customer change from the timetable for replacing underlying platforms.
Sundaram also cautioned against allowing coexistence to become permanent complexity. Each interim state, he said, needs a defined target, an owner and a retirement plan. For Sundaram, customer outcomes provide another test. He pointed to the time needed to approve a loan and abandoned journeys. “If the customer can’t feel the transformation,” he said, the bank may have focused its modernisation on the wrong layer. The same logic applies to data. A fraud model may still be technically sophisticated, but its usefulness drops if information reaches it too late to influence the transaction. In production, speed has to coexist with reliability, reversibility and control.
Production AI shifts attention beyond the model
Production AI depends on controls, reliable data and operating processes that can withstand banking conditions. Sundaram described a tiered approach to autonomy, where low-risk, high-volume uses such as fraud scoring, transaction monitoring and anomaly detection can support greater automation because the bank can measure and reverse them, while higher-impact decisions retain human accountability.
He also distinguished between a model that works in a demonstration and a system that can withstand banking conditions. Sundaram stressed that production systems have to perform reliably across thousands of transactions and edge cases under regulatory scrutiny. That requires identity controls, auditability and fallback mechanisms. He also argued that the model itself can be the least expensive part of an AI deployment, while orchestration, observability, security and retrieval account for much of the surrounding cost. In banking terms, that means coordinating how AI works across systems, seeing what it is doing, detecting when it fails and ensuring it works with information that is sufficiently current for the decision being made.
ANEXT reaches the same production challenge from a cloud-native starting point. Oh described the bank as cloud-native from launch in 2022, with APIs, distributed architecture, resilience and cloud scaling available from the outset. Production AI still requires continuous testing, human feedback and controls that remain in place after launch. ANEXT uses evaluation sets, predefined test cases used to check model performance, and has developed feedback loops so staff can assess outputs over time. At that point, execution depends on whether teams can identify worthwhile problems and move AI use cases into production without weakening the controls around them.
ANEXT brings technology closer to operating problems
ANEXT has placed technology leads in operations-heavy functions as forward-deployed engineers, effectively embedding senior technical specialists inside business teams. Oh described their role as diagnosing problems with business owners and deciding where technology or AI can produce an operational solution. ANEXT places engineers closer to the processes, decisions and operating constraints they are expected to improve.
That may help the bank test whether an AI use case solves a real operating problem, with production deployment providing the evidence that the solution works in practice. Some ideas do not reach production, Oh acknowledged, and the model gives technical specialists a larger role in developing solutions with business owners.
Core modernisation becomes a permanent banking capability
Core modernisation does not have a clean end state in either ING’s or EastWest’s approach. Tonella rejected a stop-and-start approach to core migration, describing modernisation instead as “a permanent journey”. Sundaram similarly argued that modernisation should become a permanent organisational ability that happens iteratively.
The way forward is to make change itself governable. Banks will need clear ownership of interim architectures, reusable controls for identity and data access, and production standards that can be applied consistently as new AI use cases emerge. The stronger institutions will be those that can absorb new capabilities without multiplying exceptions, duplicating control frameworks or allowing temporary architecture to harden into permanent complexity.
Stablecoins are raising pressure on bank payments and deposits, pushing ANZ and UOB to focus less on the form of digital money and more on client liquidity, settlement, asset distribution and interoperability.
Stablecoins are raising pressure on bank payments and deposits, pushing ANZ and UOB to focus less on the form of digital money and more on client liquidity, settlement, asset distribution and interoperability.
Anouska Ladds of Mastercard and Johnson Poh of Singapore’s Infocomm Media Development Authority argue that AI agents will scale only when financial institutions can govern, trace and supervise their actions inside live operations.
As wallets, instant payments and embedded finance reshape how customers interact with payments across Asia Pacific, card issuance is often assumed to be losing relevance. David Lim of Alliance Bank Malaysia, Anshul Sabherwal of Standard…
Customers in Malaysia increasingly expect international payments to be as seamless as local transfers. At a closed-door roundtable convened by TAB Global with Visa, participants examined how banks are adapting to these expectations while…
