The Role of AI-Driven Large Transaction Models in Transforming Payment Security

As the digital economy grows, so does the volume—and complexity—of digital payments. With billions of transactions now flowing daily through networks, banks, and FinTech platforms, the financial world is facing increasingly sophisticated fraud threats, according to PYMNTS. But at the same time, it is also discovering powerful new tools to combat them. Among the most promising of these innovations are Large Transaction Models (LTMs)—AI-powered systems designed to secure payment flows and enhance financial operations.

A New Layer of AI-Driven Protection

LTMs are built on transformer-based AI architectures, the same technology behind large language models like GPT-4. Just as LLMs understand and generate human language, LTMs are trained to comprehend the «language» of financial transactions.

Wolfgang Berner, co-founder and CPO of Hawk, explained the concept in a conversation with PYMNTS: “The core idea is we treat transactions as sentences, teaching the transformer model the language and grammar of transactions, similar to how large language models like GPT-4 are trained on the text of the web. And by doing that, it develops a very good understanding of the transactions, how transactions relate to each other, and what is genuine or possibly suspicious with them.”

These models process billions of historical transaction records to learn behavioral patterns, recognize anomalies, and make real-time predictions—often detecting fraud attempts that traditional machine learning (ML) models would miss.

The Evolution Beyond Traditional Machine Learning

Conventional ML models have long been used to combat payment fraud, relying on defined features such as BIN numbers or ZIP codes. While effective to an extent, these models are limited—they often require separate, task-specific algorithms for fraud detection, authorization, and dispute resolution.

LTMs, by contrast, can tackle multiple problems within a single architecture. Their ability to «embed» transactions in a vector space allows for deep semantic analysis. Transactions from the same issuer or those sharing characteristics—like an email address—are naturally grouped, making fraud patterns more visible and precise.

Staying Ahead in the Cyber Arms Race

Crucially, LTMs adapt over time. As fraud tactics evolve, so too do the models, making them increasingly effective.

“It’s become harder to monitor all the various ways that fraudsters attack businesses,” said Eric Frankovic, general manager of business payments at WEX. This sentiment is echoed across the industry, as payment providers seek to balance tight fraud detection with a smooth customer experience.

“It is essentially an adversarial game; criminals are out to make money, and the [business] community needs to curtail that activity,” said Michael Shearer, Chief Solutions Officer at Hawk: “What’s different now is that both sides are armed with some really impressive technology.”

More Than Just Fraud Prevention

While LTMs are making major strides in fraud detection, their potential goes much further. These models are starting to streamline operations in areas like compliance, risk management, and customer experience.

By automating large parts of financial workflows, LTMs help banks and institutions handle growing volumes of data without sacrificing accuracy or responsiveness. From credit scoring and strategic planning to customer service automation, the applications for LTMs are expanding rapidly.

A recent PYMNTS Intelligence report, “Leveraging AI and ML to Thwart Scammers,” produced in collaboration with Hawk, highlights how businesses can use AI not only to secure their systems but also to gain strategic advantages in a highly competitive financial landscape.

As digital payments continue to accelerate, tools like Large Transaction Models represent a critical shift—transforming AI from a back-office experiment into a frontline defense system. In the words of Hawk’s Wolfgang Berner, these models are teaching systems to “understand” transactions like language—and that understanding may be the key to outsmarting financial crime.