Anti-money laundering for banks

The constant influx of transactions presents a significant challenge, requiring extensive sifting. Consequently, this demand has given rise to a market for anti-money laundering products. Traditionally, such products relied on human intervention to manually input specific rules into the system for flagging purposes. For example, transactions exceeding ten thousand dollars were flagged. However, this approach often resulted in either an excessive number of flagged activities or insufficient detection.

Now, Google Cloud introduces a superior solution. Alphabet's Cloud business has recently unveiled a new tool that eliminates the need for human-generated rules. According to the company, this tool offers enhanced accuracy. To delve deeper into this topic, we have Dylan Tokar, a reporter from the Wall Street Journal, who has extensively researched the anti-money laundering program market. Dylan, please provide us with further insights into this development

There are numerous AML transaction monitoring, also known as surveillance programs, already available in the market. These programs are considered essential for financial institutions, as they are mandated by regulators to screen customer transactions and activities to detect any suspicious or illegal behavior. Many existing products in this domain claim to utilize artificial intelligence in some capacity, including the new offering from Google Cloud.

Google Cloud aims to differentiate itself by eliminating the reliance on manually defined rules or rule-based programming. Unlike most existing programs that require human input to define parameters and rules for detecting certain activities, Google Cloud seeks to address the challenges posed by the increasingly complex nature of AML processes and vast amounts of data that banks need to analyze.

By minimizing human input, Google Cloud intends to mitigate the problem of false positives, which occur when rule-based systems generate numerous alerts even for legitimate transactions. Each alert must be investigated by a human, resulting in inefficiencies. Google Cloud's approach involves leveraging machine learning to reduce the number of alerts generated and enhance accuracy. The system will autonomously determine which transactions require investigation, resulting in more precise and reliable outcomes, ultimately reducing the overall noise and improving efficiency.

How does Google Cloud's tool differ from others that also incorporate AI in their products? The key difference lies in the deployment and process. While other companies typically rely on manually defined rules or rules-based inputs, Google Cloud argues that their strength lies in their more robust machine learning models and algorithms. They take a different approach by training their API and machine learning model using three years of existing transaction alert data, eliminating the need for manually inputted rules. This allows their AI to proactively identify and manage alerts, rather than waiting until a problem has already occurred.

The response to this tool has been positive so far. Google Cloud has disclosed a limited number of clients, with HSBC being one of the notable ones. HSBC partnered with Google Cloud to develop the tool, as they were seeking to improve their anti-money laundering (AML) screening software, given their history of regulatory issues in that area. According to HSBC, the tool has been successful in reducing false positives by 60% and increasing true positives by two to four times. However, the ultimate adoption and success of the tool, given the immense task of screening billions of transactions, as well as the response from regulators, remain to be seen. It is worth noting that there are potential risks associated with delegating decision-making to AI systems in this context.

"How does a machine learning model or algorithm determine what constitutes a risky transaction? And what are the consequences if it fails to screen a transaction that should have been flagged?