The argument for core modernisation has moved on. Large banks increasingly accept that ageing infrastructure, fragmented data and tightly coupled systems constrain how quickly they can change. The harder question is how to modernise systems supporting millions of customers and transactions without creating unacceptable cost, operational risk and disruption.

Banks are already moving away from one-off core replacement towards more progressive and continuous modernisation, breaking transformation into smaller domains to reduce execution risk and bring forward business value. Artificial intelligence (AI) is adding another dimension: it requires modern infrastructure and reliable data to operate effectively, but may itself help reduce some of the implementation and upgrade effort required to build those foundations.

Takis Spiliopoulos, Chief Executive Officer, Temenos, said banks are becoming more disciplined about where AI produces measurable value. At the same time, they are exploring whether the technology can reduce the implementation burden that has historically made core transformation difficult.

The more consequential question is therefore not simply how much AI banks can put into the core. It is whether AI can change the economics of modernising it.

AI faces a harder test on economics and accountability

The industry's AI conversation is becoming more disciplined. Spiliopoulos said some banks had consumed what they expected to be a full-year generative AI token budget within the first quarter or first half, without yet seeing comparable tangible benefits. That is shifting the discussion from how many AI use cases banks can develop towards what they cost and what measurable value they produce. Governance becomes more difficult as AI moves from employee copilots and operational support towards autonomous agents.

Spiliopoulos used payments to illustrate the problem. If an autonomous agent fails to execute a payment, the bank cannot simply attribute the failure to the agent. Responsibility ultimately has to sit somewhere, with the bank, technology provider, systems integrator, operator or potentially an insurer. He said unresolved liability would constrain production-level applications, “definitely not on the core”.

Autonomy does not eliminate accountability. Spiliopoulos was similarly cautious about AI's current revenue contribution. He said there is relatively little scaled incremental revenue among bank clients today, with many applications still at proof-of-concept stage. The clearest measurable returns are instead emerging from end-to-end internal processes, particularly operations, where banks can quantify cost savings and reductions in manual intervention or processing time.

That makes AI's potential role in reducing the cost and effort of modernisation particularly relevant.

AI could alter implementation economics

Core banking software itself can represent only part of the economic burden of transformation.

Spiliopoulos illustrated the problem with an example he said he hears from bank executives: a software licence costing $2 million can become difficult to justify if implementation costs $50 million.

The actual numbers will vary materially between projects, but integration, migration, configuration, testing, customisation and organisational change can make implementation much more consequential than the underlying software licence.

Spiliopoulos said reducing that effort with AI could encourage more banks to undertake modernisation.

Upgrades present the same problem. Banks can postpone them because of the disruption and resources involved, but falling progressively behind makes it harder to access new capabilities and can increase the difficulty of subsequent modernisation.

Spiliopoulos posed another example: if an upgrade that takes nine months today could eventually be reduced to three using AI-assisted tools, the calculation around keeping platforms current changes.

Temenos is developing agents intended to assist with implementation and upgrades, although Spiliopoulos said they remain at proof-of-concept stage and are “not yet enterprise grade”.

Temenos had already identified installation, operation and upgrades as priorities in its 2026 AI product strategy, including the use of agents to address integration, versioning and operational complexity.

AI-assisted implementation has therefore not yet been demonstrated at enterprise scale. But it addresses one of the persistent barriers to core modernisation: banks may understand why they need to modernise and still struggle to justify the time, cost and execution risk involved.

Progressive modernisation reduces concentration of risk

The implementation problem is also changing how banks approach the core itself. Spiliopoulos said large institutions are increasingly breaking modernisation into smaller business and technology domains rather than attempting to migrate much of the bank at once.

A bank might modernise retail deposits first, followed by lending and corporate banking. An international institution could start with overseas subsidiaries before tackling its domestic franchise.

Spiliopoulos said a specific business line could potentially begin seeing value within six to nine months, compared with perhaps two years for a broader core transformation.

That approach is also appearing in Temenos's product architecture. In May, it launched standalone composable retail deposits and lending capabilities designed to integrate with existing bank technology through application programming interfaces and event-driven connections, allowing individual domains to be modernised separately.

Progressive modernisation reduces the amount of operational risk concentrated into a single cutover and brings forward the point at which individual businesses can start realising value. It also supports a more continuous model of transformation in which banks progressively change their underlying architecture as business requirements, technology and regulation evolve.

Spiliopoulos said it is also making banks more open to moving workloads to cloud, which they increasingly see as providing greater flexibility. Software-as-a-service adoption is more uneven, with smaller and digital banks generally moving faster while larger traditional institutions remain more cautious.

Vietnam illustrates how quickly those requirements are changing in some Asian markets. Spiliopoulos said Vietnamese banks are moving increasingly towards cloud, although software-as-a-service adoption remains more constrained. More significantly, he said the intensity with which Vietnamese banks challenge technology providers can generate ideas and capabilities that subsequently apply elsewhere in ASEAN. “They keep challenging us. They're ambitious, fast,” he said.

Rising incomes and rapid customer growth are also changing what Vietnamese banks need from their platforms. Spiliopoulos pointed to expanding mass-affluent and wealth opportunities as customers develop more sophisticated financial needs.

Temenos completed its acquisition of Swiss wealth technology company additiv on 17 July. The company said the acquisition strengthens its wealth proposition, particularly for mass-affluent customers, and adds AI-powered orchestration across digital and assisted customer journeys.

Vietnam, in Spiliopoulos's assessment, is therefore becoming an innovation market from which capabilities can be taken elsewhere in the region, rather than simply a destination for established banking technology.

Modernisation is both prerequisite and beneficiary

There is a paradox in AI-driven banking. AI requires the very modernisation many banks are still trying to complete. More sophisticated models and agents depend on reliable data, modern infrastructure and architectures capable of supporting more frequent change. Spiliopoulos said banks cannot move towards an AI-enabled operating model without those foundations.

But AI may also become one of the tools that makes those foundations easier to build. The near-term evidence remains stronger around operational efficiency than revenue generation. Accountability for autonomous AI remains unresolved, and the agents Temenos is developing to support implementation and upgrades are not yet enterprise grade.

Banks already know they need to modernise. The persistent obstacle has been doing so without destabilising the institution or allowing implementation cost, duration and complexity to overwhelm the business case.

If AI can materially shorten implementation and upgrade cycles, it could change that calculation—not whether banks need to modernise, but whether the cost, time and risk of doing so can finally become manageable.