How AI is helping Australia’s biggest bank stop domestic violence perpetrators

We have long-known domestic violence perpetrators carry out abuse in sophisticated ways – and as technology has become more sophisticated, they have learned how to weaponise the digital world more effectively. Over the last decade, banks have seen a steep rise in abuse through bank transactions. 

This is where a perpetrator transfers a minimal amount – as little as one cent – to their partner or former partner and uses the description section to harass or send a threat. The messages vary from the subtle (‘Have you taken out the dogs today?’) to the sinister (‘I know where you live.’) but the impact is the same: it creates an ‘omnipresence’ that makes a  person feel like it is impossible to escape their perpetrator.

In 2020, CommBank discovered the problem and became the first bank in the country to start blocking digital payment transactions that included harassing, threatening or offensive messages. Since then, they have blocked more than one million messages across the country. And, now, the bank has developed an artificial intelligence model to more effectively combat this form of technology-facilitated abuse.

On Wednesday morning, CommBank’s Head of Customer Vulnerability, Caroline Wall, explained this artificial intelligence model – which the company has been refining since 2021 – is built on a “relationship-based model” that doesn’t just look at an abusive payment description, but the abuse in context.

“The model itself uses a mixture of machine-learning, natural language processing, and some pre-trained large language models to analyse not just the individual meaning of the description, but to look at the relationship between the perpetrator and victim. So it looks at long-term patterns of systemic abuse,” Wall said. “We look at things like sexual content, we look at intent, we look at the emotive nature of the language, we also then look at patterns of behaviour.”

“Today, we’ll be making the code that we developed available to any bank or client so that they can also monitor their transactions and stop abusive messages.”

Once a case is identified, it is monitored for one to three months and given a score of the likelihood of abuse. High-risk cases are then sent to a trauma-informed specialist team, who reach out to the victim-survivor to offer security and support.

PayID features can be removed, safe accounts can be set up, warning letters can be sent to abusers and, in extreme circumstances, the bank can cease their banking relationship with an abuser. But every response is different and tailored to the individual victim-survivor.

CommBank’s Group Executive, Human Resources, Sian Lewis, announced that the bank will be making the source code for the AI model available to their competitors and clients, in a wider effort to reduce technology-facilitated abuse and end the scourge of domestic and family violence in this country.

“This isn’t a partisan issue. This is not a competitive issue for us, either. It is about hopefully being one of the leaders in the corporate charge to stamp out financial abuse and domestic and family violence,” Lewis said. “And so today, we’ll be making the code that we developed available to any bank or client so that they can also monitor their transactions and stop abusive messages.”

The model and source code will be available this week through CommBank’s partnership with H20.ai on GitHub, the world’s largest platform for hosting source code. The model was built by CommBank and the source code was developed in partnership with the bank’s exclusive partner and global AI leader, H20.ai.

CommBank’s commitment to helping end financial abuse began after former Australian of the Year Rosie Batty called on the public and private sector to play their part in creating solutions in this space. Since then, among other initiatives, the bank established the Next Chapter Financial Independence Hub, which has helped more than 5,000 victim-survivors through financial coaching services and broader support.