Banks Narrow Fraud Technology Divide but Deliver Uneven Results

 | 

Highlights

The research underlying September’s “Scams Get Personal: Why Banks Pay More to Fight Fraud” shows behavioral analytics use among these institutions rising from 74% in 2025 to 100% in 2026. Machine learning and AI adoption increased from 77% to 96%. Large banks reported universal use of both technologies in both years.

Earlier purchases can supply context for evaluating a transaction that appears unusual in isolation. Whether that information supports approval depends on the institution’s controls and the other risk signals available.

In PYMNTS’ coverage, Nilesh Dusane, global head of institutional payments at AWS, described payment models that analyze transaction sequences. A plane ticket followed by hotel and restaurant purchases, for example, supplies context for evaluating subsequent spending.

The approach illustrates how institutions using the same technology category could reach different decisions. The information available to a model and its ability to interpret relationships among transactions can affect its assessment. The coverage describes the capability without supplying comparative fraud loss results.

Customer history also has limits. A familiar device or valid login cannot establish whether a customer is following a scammer’s instructions. Customer-manipulation scams represented 22.4% of fraudulent dollar value in the bank research, up from 12% in 2025. Banks must account for payments initiated by deceived customers as well as transactions started by intruders.

Checking Before Release

Recipient verification adds another source of information before a transfer proceeds. As reporting by PYMNTS, payment provider TerraPay connected its verification service with J.P. Morgan’s Liink network, enabling institutions to validate bank account and digital wallet details for cross-border payments.

The service targets incorrect destination information, including errors that can cause failed transfers. Its introduction illustrates how providers are placing additional checks before funds leave, although the announcement supplies no measured reduction in fraud losses.

Separate PYMNTS Intelligence research associates real-time verification with faster detection. October’s “Fast Money Needs Faster Fraud Checks” found that 60% of firms verifying account ownership in real time detected suspected fraud within minutes or sooner, compared with 38% without that verification.

The survey covered 150 executives across seven industries, rather than financial institutions alone. Its results provide a comparison between verification practices and detection speed without establishing that verification caused the difference.

Additional information can improve a decision, while additional review can hold up a payment. In the October research, 71% of firms cited manual review bottlenecks as a leading cause of false declines or payment delays. Overly conservative rules followed at 69%.

These findings give banks a reason to examine what follows detection. Sending more transactions to employees can consume resources and inconvenience customers even when the detection system operates as intended.

The bank research demonstrates why financial performance requires several measures. Neobanks and FinTechs reported a lower fraud loss rate in 2026, yet estimated dollar losses and prevention spending rose sharply. Small subgroup samples warrant directional comparisons, and the study doesn’t establish which practices caused the changes.

As banks consider further investment, customer information, verification timing and review capacity offer specific areas for improvement. Their value will depend on whether institutions can reduce losses and unnecessary intervention while allowing legitimate payments to proceed.