Responsible use of models

Why does TMNL use models to monitor transactions?

TMNL’s core task is to combine transaction data from different banks and use them to make meaningful connections. These connections provide new insights and signals with regard to potential money laundering or terrorism financing activities. This requires the use of models for data analysis, because they are an effective tool for differentiating possible criminal cash flows from normal transactions

Banks have been monitoring transactions for many years. What is TMNL doing differently?

As gatekeepers of the financial system, banks are required by law to monitor all of their customers’ payment transactions. To do that, banks have long used models to detect potential signals of money laundering or terrorism financing.

But criminals often use multiple banks to spread out their transaction flows. TMNL utilises the best practices from the participating banks, combines the transaction data for joint monitoring, and in so doing gains a comprehensive insight into possible unusual transaction patterns. This allows TMNL to reveal potential criminal cash flows that aren’t visible for individual banks

TMNL uses models to detect potentially unusual transactions. Does it also use automated decision-making?

The decision-making process is not automated. The models only combine statistics, technology and data. First, experts in the field of money laundering detection and data science determine which data and accounting rules need to be utilised. Then the models differentiate between normal transactions and those that may be unusual. When TMNL identifies a potentially unusual transaction, it sends an alert to the bank. The bank employee then investigates the alerts for potential criminal activity.

What does the bank do with an alert for a potentially unusual transaction?

When a bank receives an alert from TMNL, it uses the customer information available to investigate whether the transactions display the characteristics of potential money laundering or terrorism financing. If necessary, the bank then contacts the customer. Depending on the results of the investigation, the bank may report the transaction to the Netherlands Financial Intelligence Unit.

The media frequently reports on examples of models that can lead to discrimination. How does TMNL safeguard against that?

TMNL uses a Model Risk Framework based on De Nederlandsche Bank’s SAFEST principles. This framework provides a single context for the responsible use of Artificial Intelligence and models. An external auditor also regularly reviews the structure, existence and function of the models, with the assistance of a special committee of external independent experts and representatives of the member banks.