Be the Bank That Scales With AI in 2027

The core problem for banks adopting AI is not the AI itself, but the legacy banking software. MIT’s Project NANDA shows 95% of generative AI pilots fail to deliver returns. This failure stems from the 80% of work needed for production, which involves data engineering, governance, and integration. Existing core banking systems, built for human clicks, cannot handle the demands of AI agents querying data and calling APIs directly.

This creates an opportunity for new players in Southeast Asia. Salmon in the Philippines, for example, scaled to millions of customers and grew its loan portfolio by 648% in twelve months using an AI-native architecture. FairMoney manages over 24 million accounts and processes 8,000 loan applications daily. Regulators like OJK in Indonesia and BSP in the Philippines are issuing AI governance guidance, pushing institutions toward auditable logic and defined points of control.

The test for incumbent banks in the region is whether they can adapt their core systems by 2027. Most cannot today. The challenge for them is to implement a governed data foundation, complete API coverage, and a safe execution layer. This will determine which institutions can truly scale with AI and meet evolving regulatory expectations.