
AI success is prompting leaders across the region to re-examine long-standing operating habits. The shift is not merely about adding new tools; it is about reshaping how work gets done, where data processes live, and which infrastructure can keep pace with evolving demands.
Adoption gaps reveal hidden inefficiencies
A recent McKinsey State of AI report shows that 89% of firms regularly employ artificial intelligence in at least one function, yet only 37% say it lifts enterprise-level earnings. Yet widespread adoption has not translated into equally widespread returns.
When a system excels at a single step, its impact stalls if surrounding procedures cannot absorb the speed gain. Faster analysis means little if decisions remain delayed, staff lack authority, or existing systems cannot handle additional workloads.
In the financial-crime arena, the technology can flag suspicious transactions and rank alerts. However, if investigators must still sift through every notification manually, the net workload may rise rather than fall.
Singapore’s banking community is experimenting with a new approach. Launched in September 2026, the Institute of Banking and Finance‘s AI Workforce Co-Lab brings together more than 80,000 staff from 23 institutions to prototype job redesign and skill-building that align automation with human judgment.
For regional players, the challenge deepens. Different markets impose distinct regulatory rules, data-residency mandates, and language requirements. A workflow that thrives in one jurisdiction may need separate controls, hosting choices, or oversight in another.
Businesses must decide which activities can run unattended, when staff should intervene, and who bears ultimate responsibility. Employees also need the ability to read system outputs, question recommendations, and manage outliers.
Rethinking where workloads run
During trial phases, the goal is simply to prove feasibility. Once a solution moves to production, the underlying environment becomes a strategic choice. A customer-facing app that demands instant replies may favor low-latency, high-availability settings, while a platform handling sensitive financial data must prioritize security and compliance.
Enterprises that span Southeast Asia often juggle mainframes for core transactions, public clouds for client portals, and private setups for confidential workloads. Selecting the right placement therefore requires a granular assessment of each use case, not a blanket cloud-or-on-premises rule.
These placement decisions are not static. An application initially deployed in a public cloud for experimentation may require a different architecture as usage grows, costs change, or regulations evolve.
For instance, instead of replacing an entire legacy system, a firm might modernize its API layer, enabling the new analytics engine to access needed data without a full rebuild. The technology can even assist by scanning old codebases, mapping dependencies, and suggesting refactoring steps, though human experts must still validate any change.
The underlying infrastructure must be able to support the demands of AI workloads, including low latency and high availability.
