The Agentic AI paradox in Singapore fintech: Why adoption is outrunning infrastructure

Financial institutions (FIs) across Singapore are moving quickly to adopt AI, with agentic systems increasingly at the center of that push. A new study from Confluent finds that three in four Singapore IT leaders are already deploying or piloting agentic AI, ahead of most global markets. Yet the same study finds that 78% of those leaders don’t have the real-time data infrastructure to support it. Infrastructure gaps and weak AI governance are the two biggest reasons agentic AI projects stall or get shelved entirely.

For fintechs, this matters more than it sounds. Many are chasing hyper-personalization: real-time offers, tailored pricing, AI-driven advice. But that promise hasn’t reached customers yet.  A recent industry discussion at Money20/20 Asia found that just 17% of Singapore consumers are satisfied with bank personalization. Closing that gap depends on fixing the same two problems: the data feeding these systems, and the governance controlling what they’re allowed to do.

Starting with the data. 78% of Singapore’s IT leaders say they don’t have the real-time infrastructure to run AI properly, well above the 72% global average. Fragmented data ownership (73%) and a shortage of AI skills (also 73%) round out the list. The result: 73% of agentic AI projects in Singapore have stalled, and half have been abandoned outright.

Here’s the odd part. A separate report from core-banking vendor Finastra found that 71% of Singapore financial institutions rate their own infrastructure ahead of global peers, the highest score of any market surveyed. Two studies, two very different pictures. One says infrastructure is the top complaint. The other says infrastructure is a strength. Both can’t be fully right and the gap between the two is probably the more honest read: institutions feel ready on paper, but the moment they push AI from a pilot into something that acts on its own, the cracks show. Confidence and readiness are not the same thing, and agentic AI is what’s exposing the difference.

Then there’s governance. Weak AI governance is happening inside companies, unclear policies and low visibility into what agents are doing to name the least. This is where Singapore’s MAS has gotten ahead. The regulator has recently published a whitepaper called the SAFR (Safeguards for Agentic Finance at Runtime) together with leading financial institutions. It gives fintechs a shared set of guardrails to strengthen their AI governance as agent adoption grows.

The idea is simple: before an AI agent can make a payment, review a client document, or take any other action, SAFR checks it against a set of rules. If the action breaks those rules, it gets blocked or sent to a human to approve. Every decision gets logged. Industry members are already testing it in payments, treasury operations, and wealth management document review. That’s a genuinely useful set of guardrails. SAFR won’t fix the data infrastructure problem, but it removes one real excuse for holding back on agentic AI.

Put the two pieces together and the message for fintechs is simple. Personalization promises don’t mean much without real-time data behind them. And agentic AI projects won’t get past compliance without something like SAFR wrapped around them. The vendors and fintechs that spend the next 12 to 18 months fixing both are the ones that will get to build the hyper-personalized products they keep pitching. Everyone else is still selling a slide.