Asia-Pacific banks want AI to fight fraud, but their biggest hurdle is privacy, not cost

By Asian Business Review · October 18, 2026.

More than 60% of banking leaders in Asia-Pacific (APAC) name fraud and scam detection as the area where generative AI can most improve their operations over the next 12 months, according to a survey by analytics software company FICO. Credit underwriting decisions come second (21%), followed by customer-facing digital banking services (15%) and regulatory compliance monitoring (3%).

The main barrier to adoption is data privacy and security, cited by 42%. Opacity and regulatory risk weigh less: model transparency and explainability, and regulatory and compliance risks, each account for 21%. Only 16% point to compute or infrastructure costs.

Some 77% consider it "extremely important" that banking AI tools be purpose-built for the industry, and the remaining 23% see it as "somewhat important". On timelines, 46% expect to deploy task-specific models (for fraud, compliance and underwriting) within one to two years, 26% within three to five years, and another 26% say they have already deployed them. Beyond fraud, 64% believe credit underwriting and lending decisions are the back-office processes that would benefit most.

The survey was conducted in August 2026 at FICO's Banking Leaders Forum in Bali among 51 executives from APAC banks. Dattu Kompella, FICO's managing director for APAC, argues that banks need models that explain their decisions, are auditable and comply with regulation.

**Our take.** The data needs to be put in proportion before anyone celebrates it. These are 51 responses, gathered at a forum hosted by a vendor that sells exactly this kind of product. They work as a gauge of intentions, not as proof of real-world impact. And "already deployed" for 26% doesn't say whether it works, only that it exists.

Even so, the signal is consistent. Fraud is the ideal use case for AI in banking: high volume, shifting patterns and a clear cost for every mistake. The threat is being automated too. Scammers already use AI and deepfakes, so defenses have to move just as fast. Broadly, this race between attack and defense is what pushes banks to invest more urgently here than in other uses.

Most interesting is what the ranking of obstacles reveals. Compute worries 16% while data worries 42%, which suggests the technology is now seen as accessible and the bottleneck has shifted to trust: where customer data lives, who sees it and how a decision is audited. It is the same logic of sovereignty and control we see in open-weight models. The preference for industry-specific models over general-purpose ones points the same way: smaller, auditable and governable.

For jobs, the nuance matters. The 64% who single out lending and underwriting as the back-office processes with most to gain are pointing at exactly the administrative ground that, in our series on banking and insurance, we see as most exposed. In the short term, that means fewer administrative staff and more data, risk and compliance roles, with a transition that won't be painless for those doing that work today.

In the long term, the picture is more constructive. A financial system that curbs fraud at scale mainly protects those who lose most to it: older people, small businesses and users with fewer resources. And if the models are also explainable and auditable, AI can widen access to credit without opacity. That is the condition: explainability must be a verifiable requirement, not a slogan.