Singapore bank turns on AI to fight vice activities

OCBC Bank has been trialling machine learning technology to analyse corporate bank transaction data and identify suspicious activities, specifically, money laundering, at four times the accuracy rate.

OCBC Bank is trialling the use of artificial intelligence (AI) and machine learning tools to improve its ability to accurately identify suspicious activities, specifically, money laundering.

The Singapore bank said it ran the software through one year's worth of corporate banking transaction data and reduced, by 35 percent, the number of alerts that did not require further review. It also improved the accuracy rate of identifying suspicious transactions four-fold by categorising flagged activities according to risk levels.

Cashless cannot be the face for Singapore smart nation success

Increased emphasis on building a cashless society to reflect country's success as a smart nation is misplaced, when the importance of getting the fundamentals right is overlooked.

Read More

OCBC said its fintech business unit, The Open Vault, conducted the proof-of-concept for the software earlier in the year with another fintech company, ThetaRay. The bank said the software now was in an "extended proof-of-concept and pre-implementation phase", which would involve advanced testing with additional test data.

It said this would enable the bank to further evaluate the efficacy, security, and robustness of the application. Once successfully completed, OCBC said it was targeting to implement the tool in the second quarter of 2018 and have it run alongside its existing transaction monitoring system.

With the new software, the Singapore bank hoped to improve efficiencies in the way it currently monitored vice activities.

It explained that an internal anti-money laundering compliance analyst had to review hundreds of potentially suspicious transactions flagged by the company's transaction monitoring system. These would have been highlighted upon meeting one of several pre-set rules, such as sudden large transfers between accounts.

OCBC noted that such rules-based model generated multiple alerts that its analysts had to manually scour to determine if a transaction warranted further review for potential financial criminal activities. It said this could take days or up to a week to review, depending on the complexity of the transaction.

It also would not account for suspicious activities that the system failed to detect if these transactions did not "meet" any of the pre-determined rules.

According to OCBC, the AI software used an algorithm that was not dependent on a finite set of rules. Rather than analyse transactions as separate, individual activity, the application was able to analyse broad parameters--encompassing products, customers, and risks--to detect anomalies in transaction behaviour. It also could look at diverse data sources to make contextual data analysis, the bank said.

"Furthermore, the software is dynamic and is able to learn from or adjust to changes in transaction patterns over time, allowing it to flag suspicious transactions with better precision as well as discovering new patterns for smarter future detection," OCBC said in a statement Tuesday.

This would reduce the number of transactions that needed further review. The software also would be able to categorise alerts by levels of risk, hence, improving the accuracy of detecting suspicious transactions since OCBC's analysts would be able to prioritise alerts that carry higher risks.

Pay by Apple Face

In a separate announcement Monday, OCBC said it had added support for Apple's Face ID facial recognition, enabling iPhone X users to authenticate access using their face.

To use the feature, customers would need to activate OCBC OneLook via the bank's various mobile apps including OneWealth and Pay Anyone.