Visa is strengthening its push into real-time payments security with an enhanced version of A2A Protect, adding a unified fraud score powered by technology from Featurespace and giving banks a new tool for identifying suspicious account-to-account transfers before money leaves a customer’s account. The product update is the first in-market integration combining Visa and Featurespace technology since Visa completed its acquisition of the Cambridge-founded fraud detection company in December 2024.

The rollout places fraud scoring at the center of Visa’s strategy for the expanding account-to-account, or A2A, payments market. Unlike a conventional card transaction, an A2A payment transfers funds directly between bank accounts through payment infrastructure that can settle in seconds or near real time. That speed improves convenience and liquidity for consumers and businesses, but it also shortens the window in which banks can identify suspicious behavior and intervene before funds become difficult to recover.

Visa said the enhanced A2A Protect generates real-time risk insights through a combination of advanced artificial intelligence, transfer-learning techniques and network intelligence. The system is designed to provide useful fraud signals from the beginning of a deployment rather than requiring months of historical transactions from an individual bank before a model becomes effective. Financial institutions can also opt into additional network-level intelligence designed to identify threats that may be difficult to see from the transaction history of a single institution.

The unified score is important because fraud detection increasingly depends on combining multiple views of a transaction. Featurespace specializes in behavioral analytics that establish patterns of normal customer activity and identify deviations that may indicate scams, account takeover or other financial crime. Visa contributes payment-network scale and risk signals spanning a much wider ecosystem. Combining those sources potentially gives financial institutions both a granular view of individual behavior and broader intelligence about emerging fraud patterns.

Featurespace was founded in 2008 and developed technology that learns how legitimate customers normally transact rather than relying solely on static fraud rules. Visa announced its agreement to acquire the company in September 2024 and completed the transaction on Dec. 19 of that year, placing Featurespace within its Risk and Identity Solutions business. Visa said at the time that it planned to incorporate Featurespace’s technology into its existing fraud-prevention and risk-scoring products. The enhanced A2A Protect is one of the clearest commercial outcomes of that integration.

For Visa, the strategic significance extends beyond fraud prevention. The company has increasingly emphasized value-added services including authentication, cybersecurity, risk analytics, advisory services and payment processing technologies that can be used by banks and merchants independently of traditional card transactions. A2A Protect fits directly into that expansion by allowing Visa to participate economically in payment flows that may bypass card networks altogether.

That positioning has become more important as instant-payment systems gain adoption in markets around the world. Visa cited industry projections showing global A2A transactions could exceed 5.8 trillion by 2028, representing growth of roughly 160% from 2024. In the Asia-Pacific market, where Visa announced the latest enhancements on Sept. 2, the company said the region is expected to account for more than half of global consumer A2A transactions by 2028.

The scale of that growth raises a central issue for banks and payment operators: transactions must become faster without making fraud prevention weaker. Traditional fraud systems frequently depend on rules, historical patterns or post-transaction investigation. Instant transfers change the economics of that model because a payment may be completed before a manual review can take place. Prevention therefore shifts toward scoring transactions at the point of authorization and deciding, within milliseconds or seconds, whether to allow, challenge or stop them.

Visa said A2A Protect has been shown to reduce more than 50% additional fraud and reduce unnecessary fraud alerts by more than 40%. The company separately said deployments have increased fraud detection by as much as 75% during the first six months. Those performance figures are supplied by Visa and may vary by institution, transaction mix and implementation, but they highlight the commercial proposition: fraud systems have to improve detection without generating a corresponding surge in false positives.

False positives are a particularly important issue in instant payments. A fraud model that blocks too aggressively can create customer-service costs, interrupt legitimate transfers and undermine confidence in a payment product. Banks therefore evaluate fraud technology not simply on how many suspicious transactions it identifies, but on whether it can distinguish fraudulent behavior from unusual but legitimate customer activity. Behavioral modeling is intended to improve that distinction by analyzing how a transaction compares with the established behavior of the person or account involved.

A2A Protect also gives financial institutions the option to incorporate network-level signals. Visa said participating institutions can use those signals to identify emerging scam hotspots and coordinated activity across the ecosystem. That approach addresses a structural weakness in institution-by-institution fraud detection: a bank may see only one small part of a broader scam campaign, while a network-level system can potentially identify connections among transactions occurring across multiple institutions.

The model could become increasingly relevant for authorized push payment scams, in which consumers are manipulated into sending funds to criminals themselves. Those transactions can be difficult for conventional systems because the customer is properly authenticated and has technically authorized the transfer. Detecting the fraud requires understanding context — whether the amount, destination, timing or pattern is abnormal — rather than merely determining whether valid credentials were used.

Visa’s enhanced product also includes operational features aimed at making automated scoring easier for bank fraud teams to use. A2A Protect connects with existing financial-institution systems through a single application programming interface, according to the company. Alerts include plain-language explanations describing why a payment has been flagged, giving investigators more information to support an authorization decision or subsequent review.

The emphasis on explainability reflects a wider challenge in deploying machine learning across regulated financial services. Banks may be reluctant to rely on sophisticated models if investigators cannot understand why a transaction received a particular risk assessment. Fraud platforms therefore increasingly compete not only on predictive accuracy but also on how effectively they present evidence to human investigators, document decisions and fit into existing compliance processes.

Alongside the A2A Protect update, Visa disclosed that it is developing Visa Graph IQ, a complementary graph-powered capability intended to accelerate complex fraud and risk investigations. Visa said the technology is designed to help institutions identify networks of connected fraudulent behavior, uncover money-mule activity and recognize emerging threats. Graph analysis can be particularly useful in financial crime investigations because individual transactions that appear unremarkable may become suspicious when relationships among accounts, devices and counterparties are examined collectively.

The addition of an agentic investigation capability also points to the next phase of competition in financial crime software. Fraud systems have historically focused on generating alerts and risk scores. Newer platforms increasingly aim to support or automate parts of the investigation process as well — linking entities, assembling relevant evidence and helping analysts prioritize cases. If effective, those tools could reduce the time required to investigate complex fraud networks while allowing specialized staff to focus on higher-risk cases.

The stakes are especially high in Asia-Pacific. Visa’s regional announcement cited estimates that Asia-Pacific accounts for roughly 67% of the world’s $1.03 trillion in annual scam losses, with Asia recording about $688.42 billion of scam-related losses in 2024. Those estimates come from the Global Anti-Scam Alliance and illustrate why financial institutions, regulators and payment providers are placing greater emphasis on stopping fraudulent transfers before authorization rather than relying primarily on reimbursement or recovery after funds have moved.

For banks, integrating a third-party scoring layer involves a series of trade-offs. The technology must deliver enough incremental fraud detection to justify its cost and technical complexity, operate with sufficiently low latency for real-time payments, and produce scores that complement rather than conflict with existing internal controls. Institutions also have to decide how extensively they are willing to participate in broader intelligence-sharing systems and how those systems fit with privacy, security and regulatory obligations in individual markets.

Visa’s architecture attempts to address part of that deployment burden by offering a single API and models intended to deliver useful results without requiring lengthy institution-specific training periods. That approach may appeal particularly to smaller banks and fintechs that lack the transaction volumes or internal data-science resources of the largest global financial institutions. At the same time, larger banks may value the network perspective as an additional signal layered on top of their own mature fraud systems.

The rollout also strengthens Visa’s position in a competitive market for financial-crime technology. Banks can choose among specialist fraud vendors, core banking providers, payment processors, internal machine-learning systems and products from global payment networks. Visa’s advantage is the breadth of payment intelligence it can potentially combine with specialized technology such as Featurespace. Its challenge is demonstrating that those broader signals produce measurable improvements over the increasingly sophisticated models already deployed by major banks.

Featurespace’s integration is therefore strategically important beyond the A2A Protect product itself. When Visa completed the acquisition, it said the transaction would combine complementary fraud and risk-scoring tools while expanding real-time detection of sophisticated attacks without adding unnecessary friction to legitimate payments. The introduction of a unified A2A score shows Visa moving from acquisition integration toward bundled products designed for deployment across new payment rails.

For the broader fintech market, the announcement reflects a shift in where value is accumulating as payments become faster and more interoperable. Infrastructure that simply moves money is becoming increasingly standardized in many markets. Fraud prevention, identity, risk management and transaction intelligence are consequently becoming more important sources of differentiation and recurring software revenue for payment companies.

The trend also challenges the traditional distinction between card networks and account-to-account payment providers. Visa does not need an A2A transaction to run over a Visa-branded card in order to sell fraud scoring, analytics or investigation software to the bank processing it. That creates a pathway for the company to remain embedded in the economics of digital payments even as consumers and businesses adopt payment methods outside conventional card rails.

The near-term test for enhanced A2A Protect will be adoption and measurable performance at financial institutions operating real-time payment systems. Banks will look for evidence that Featurespace-powered scoring raises detection rates, reduces manual alert volumes and can be introduced without creating additional payment friction. Institutions participating in network-level intelligence will also determine how effectively Visa can detect cross-bank patterns that would otherwise remain fragmented.

Visa’s Sept. 2 announcement signals that fraud prevention is becoming a central competitive layer in the global move toward instant payments. By combining Featurespace behavioral modeling, Visa network intelligence, explainable risk scoring and emerging graph-based investigation tools, the company is seeking to position itself not only as an operator of payment networks but as a provider of security infrastructure across payment types. As A2A volumes rise, the ability to identify fraud before authorization — rather than pursue funds after settlement — is likely to become one of the most consequential technology battlegrounds in digital banking.