Manish Sharma, head of business development, partnerships and marketing at Asia Pacific, Worldline, treats card and payments modernisation primarily as a sequencing problem. Responsible for the firm's regional business, he works with banks navigating card issuing and merchant acquiring modernisation across the region. For most banks running long-established systems, the more commercially viable path, in his view, is progressive extension of what already works. "Progressive is the way banks work," he said. "There is no way they could shut off tonight and the next day they pick it up." Three elements support that position. Integration complexity makes wholesale replacement structurally difficult, modular architecture is making layered extension more sustainable, and AI is adding new ways for legacy environments to be navigated and extended. Why rebuilding is rarely a clean choice Banks operate inside dense integration environments where card systems, core banking and customer-facing applications depend on each other in ways that cannot be unwound overnight. "The integration complexity always demands that you actually progressively modernise," Sharma said. "It was always the case." That shapes how migration is sequenced. Some banks begin with debit cards while others start with merchant acquiring, depending on the institution's commercial priorities at the time. Sharma highlighted the limits of replacement as a strategic posture. "What is going to be modern today is going to become legacy in few years and maybe even faster," he said. "If you can do it right now with a fraction of a cost, would that be acceptable? Or do you really want to go and modernise everything?" The key question is whether full replacement truly addresses the bank’s immediate business needs, even if legacy systems will eventually require modernisation. How modular architecture changes what extension can do Adding new capability to legacy systems has historically been slow and expensive. Sharma argues that modular architecture is changing what is possible. He noted Worldline has redesigned its newer product, PaySuite, around microservices and an API-driven architecture. PaySuite is a payments platform designed to allow banks to modernise incrementally without fully replacing legacy systems. Its modular structure means that similar APIs can operate across both legacy environments and the new platform, making migrations more manageable. "That just made the progressive nature of the implementation project more sustainable, more doable," Sharma said. The agentic commerce example illustrates the point. Building know-your-agent capability into existing card systems is possible, but doing so inside a microservices environment that integrates with what already exists is, in Sharma's view, faster and less expensive. This is the operating logic behind the analogy Sharma returns to throughout the conversation. "You've got a house and you've got a new extra family member. Do you want to build another room or do you want to bring the whole house down and do it all over again?" Where AI makes legacy environments more workable Sharma described two specific ways AI is changing what can be done inside existing card and payments environments. The first concerns visibility. Card platforms typically accumulate code over years and sometimes decades, with customisations added and features extended in ways that are not always consistently documented. Active code, defunct code and batch dependencies sit layered inside systems that are difficult to fully map. AI is now changing what is visible inside these environments. "With AI, we actually can now finally go through the code, which are thousands and maybe even hundreds of thousands of lines, and we can actually see what is defunct or not usable anymore and what is being used," Sharma said. The same capability extends to identifying batch dependencies and points where human intervention is required. Sharma described the result as "almost like an X-ray vision" into how legacy environments actually function. The second concerns commercial outcomes inside infrastructure banks already run. Sharma highlighted two areas where AI is already producing impact. Fraud detection is the first. Conventional rule-based systems force a difficult trade-off, where tighter rules reduce fraud but also block legitimate transactions, lowering authorisation rates and constraining business volume. "The first benefit that comes is you can keep your authorisation rates high and reduce the fraud rate to become lower," he said. Credit scoring is the second. Expanding lending volumes normally means accepting a higher share of delinquent loans. Worldline is working with a partner on credit scorecards combining machine learning and AI to sharpen customer selection, so that lending growth does not come with a matching rise in delinquent loans. "When you open the gate and you want to lend more money, you are actually increasing your business, but at the same time, you're also taking up delinquent loans," Sharma said. "With AI, you can actually now select which customers to give the loans to or cards to, which customers not to." In both cases, AI is changing where the trade-off between growth and risk sits. Why explainability is becoming the working standard Sharma's position is that regulatory frameworks for AI in banking remain in development, and that vendors will need to set their own standards in the interim. "In the lack of regulatory requirements being put onto us right now, we're basically putting ourselves some regulatory requirements," he said. Worldline's approach focuses on explainability. AI-driven decisions inside its products are logged with the reasoning behind them, so that the basis for an action can be reconstructed. Sharma also pointed to the limits of debiasing. AI models are trained on historical data that reflects past institutional decisions, and removing those biases entirely is, in his view, almost impossible. The more achievable discipline is making the reasoning visible enough to be challenged. He illustrated the point through a declined transaction. A consumer rejection currently generates a reason code that requires manual lookup to interpret. AI offers the possibility of explaining the decision more directly, such as identifying a specific data mismatch, allowing a legitimate customer to address the issue. When extension makes more commercial sense than replacement Sharma acknowledges the contrary view directly. The sceptical position is that progressive extension postpones the inevitable, allowing banks to keep systems alive that should be retired in favour of fundamental rebuilds. His response is grounded in commercial discipline. Worldline itself, Sharma acknowledged, chose to build PaySuite as a modern application. His point is that the choice should be tied to a business need that cannot be met any other way, and that wholesale replacement is too often assumed to be the only credible path. For Sharma, the choice between living with legacy constraints and accepting the cost of replacement is becoming less binary, with modular architecture and AI together offering more ways to extend what banks have already built. The question worth keeping in view, in his framing, is which parts of the system genuinely need replacing, and whether the business need behind a modernisation programme can be met inside the system already in place.
