NOTO: Why AI Fraud Prevention Needs Human Interaction to Beat the ‘Tick in the Box’ Mentality

Tristan Prince from NOTO and Robert Brooker from Opus Advisory Group centered on the twin pressures forcing a revolution in fraud prevention: stringent new regulation and the immediate threat of high-velocity, AI-enabled financial crime.

Prince opened by stressing that the Economic Crime and Corporate Transparency Act (ECCTA) has introduced a provision for failure to prevent fraud which places a burden on organizations to prove that they, along with their employees and affiliates, have sufficient processes and systems in place to prevent fraud.

He presented the core challenge facing organizations: siloed technology and offered the stark example of a customer whose KYC check fails at a call center, only for their account to be emptied via an ATM shortly after. He argued that many organizations cannot connect these disparate data points across the entirety of the customer journey because different systems, such as application fraud, transaction monitoring, and biometrics solutions are not sharing crucial signals. This fragmentation means organizations cannot evidence with certainty that they have done everything possible to prevent fraud.

The need for change is being accelerated by AI-enabled fraud, NOTO warned that the sheer “velocity and the volume of fraud” that organizations are now seeing will not be stopped by heritage fraud controls and starkly illustrated the mismatch, asking how a system with a one-transaction-per-second limit can manage an AI-enabled attack hitting at a thousand transactions per second.

Prince then detailed NOTO’s strategy for helping organizations future-proof their operations which involves starting by centralizing case management so analysts can see all data in one space. Crucially, the solution involves using machine learning (ML) to make effective decisions and NOTO is a strong advocate for supervised machine learning, which Prince noted is far more effective than unsupervised ML in the long term, provided the organization has the right system for data inputs and rules in place to build a model over time.