Following a successful trial with London-based financial giant HSBC, Google has introduced Anti-Money Laundering AI (AML AI), a set of tools that the company claims are significantly more efficient than traditional rule-based methods for detecting large-scale money-laundering activities.

The UN estimates that 2% to 5% of global GDP is involved in money laundering, linked to illegal activities like human trafficking and terrorism financing. Financial institutions face significant challenges in combating this problem because of the complexities of the relevant data and technology. Existing methods, which are often based on manual rules, generate numerous false positives, leading to inefficiency and even high staff turnover stemming from frustration.

AML AI provides risk scores based on transaction data, accounts, know-your-customer (KYC) information and previous suspicious activity, which analysts review in a case management system. During the trial, HSBC experienced increased positive alerts by two to four times, along with a 60% reduction in false positives, according to HBSC and Google Cloud.

The road to AI in financial services risk management

Deploying AI as the primary means of detecting money laundering has been difficult due to regulatory and risk coverage concerns. To address these issues, Google Cloud developed AML AI to offer model governance, extensibility and secure deployment within the customer’s tenant data. This means that the product is designed to be flexible and adaptable to meet each customer’s specific needs and priorities. It is deployed securely within the customer’s environment, ensuring the data’s protection and confidentiality.

Google’s APIs offer a straightforward integration process with an adaptable data structure and the ability to customize risk typologies. The company says its goal is to make the transition easier for customers and as de-risked as possible. In an analyst briefing, Google acknowledged that completely replacing the existing rule-based system may not be the best approach initially. Instead, the company encourages customers to gradually build confidence by using the systems in tandem. This combination allows financial institutions to establish the necessary proof points and mitigate risks throughout their transition process.

Explainability is essential in AI

Google Cloud aims to address concerns about the limitations of AI by introducing an “explainability” feature in its product. Instead of solely delivering transaction alerts, the solution analyzes diverse data sets to identify high-risk retail and commercial customers. When flagging a customer, the product provides detailed information about the transactions and contextual factors contributing to the assigned high-risk score, enhancing transparency and understanding. This explainability aids investigations, assists risk managers in determining covered risks and facilitates model governance.

Phasing in AI with rule-based reinforcement

Running both AI and rule-based systems in parallel allowed HBSC to consider whether the bank needed every control it had in place or if Google’s solution satisfied those needs. These considerations were addressed in a briefing with Richard May, global head of complex investigations and group head of risk assessment at HBSC. “For example, there are still some rule-based systems that exist in our bank where we have not yet deployed this product,” he said. “There was an initial concern that, upfront, we may have to keep those systems running to give us all the risk coverage we needed. Through our evidence-based approach of deploying this, we have not seen an instance where a rule-based system had to remain in effect because the product wasn’t doing what it needed to do.”

HBSC appears to have taken a very measured approach to deployment, working with internal and external stakeholders and regulators. May said the task was met with various levels of skepticism versus support. Ultimately, the outcomes it achieved—successfully finding more financial crime—created reason enough to deploy this emerging technology to combat a decades-old problem. It came down to proving that AI could more effectively do what the incumbent systems got right most of the time.

As May stated “We don’t want criminals in our bank. Full stop.” One thing he didn’t address was that stopping these criminals with AI should have a positive effect on the industry’s low retention rate with AML staff. Providing a more sensitive and nuanced assessment framework allows risk officers to better do their jobs at the same time it gives them more structured data to support decision making.

Folding up

AML AI offers a needed solution in the financial services industry, providing improved efficiency, reduced false positives and enhanced risk detection capabilities to effectively address the complex challenges of money laundering. The phased deployment approach taken by HSBC demonstrates the careful consideration given to stakeholders and regulators, resulting in successfully adopting this emerging technology to combat money laundering.

The AML AI launch also reflects Google’s ongoing commitment to integrating AI into its products, including plans to incorporate the technology into 25 different offerings, according to the company.

Google has other notable AML AI customers, including Brazil’s Banco Bradescoand Lunar, a Denmark-based digital bank. With an expanded rollout, I look forward to seeing a reduction in the percentage of GDP allocated to the clandestine practice of money laundering.