General Bank of Canada’s Chief Risk Officer shares tangible examples of AI-driven impact that leaders in any industry can learn from
By Corinium Global Intelligence
Realizing a return on AI investments is proving notoriously difficult for many leaders. But for certain financial institutions the technology is already yielding measurable results at scale, helping to detect and prevent fraud and to fight money laundering.
Businesses in any industry can learn from these use cases, but success requires targeted, incremental adoption and strong governance, rather than chasing every new technology at once, says Adam Ennamli, Chief Risk Officer at the General Bank of Canada.
“If you start with very specific business lines it will be better and will have better returns than if you started trying to boil the ocean,” he says. “So hybrid is key: Understanding everything, having talent that can navigate between the different models and techniques is a very big success factor.”
Ennamli cites three real-world examples of organizations successfully implementing those strategies, leading to billions in fraud recovery, reduced false positives in anti-money laundering and increased detection rates translating into risk avoidance.
1. The Treasury Department prevents $4bn in fraud and recovers $1bn
The United States Treasury Department launched an AI-driven fraud detection solution, fine-tuning from pilot to full deployment over 18 months. By the second year, the system learned patterns and the results improved significantly. The program processed about 1.4 billion annual payments, screening some $6.9 trillion in transactions, and prevented $4 billion in fraud. It also recovered $1 billion in check fraud.
Ennamli says the solution was developed using supervised learning for known patterns and unsupervised learning for unknown threats. That was paired with implementing risk-based transaction training and building automated pattern recognition for payments anomalies.
“There are multiple things that you need to bring in, there is no one key solution that you will stick with,” he says. “It's very important to take a risk-based approach. We hear it a lot, but it means something very simple. It means tackle the areas where the gains are highest.”
This case underscores that even government agencies, which often face stringent regulatory requirements, can successfully implement sophisticated AI systems, he says.
“If the government can do this, the private sector can too,” he says.
2. A US bank reduces false positives in anti-money laundering by 95%
A major US bank slashed the costly burden of false positives in its anti-money laundering system by 95%, freeing up investigators to focus on genuine suspicious activity rather than spending resources on chasing benign alerts.
“The longer false positive lives the more money spent on it,” Ennamli says. “If it's a real positive, the better you are managing your resources.”
Ennamli says by analyzing transaction patterns and customer behavior, the bank reduced false positives by about 95%. He highlights this case to illustrate the potential efficiency gains when quality data is properly aligned with risk signals.
"Data is the basics of everything so it speaks to it mainly like you can't just Start working with bad quality data. You will get bad quality outcomes."
3. Graph neural networks boost detection rates for compromised cards
Ennamli points to a leading global payments company that built in-house graph neural networks to analyze merchant relationships. Their approach led to a 20% increase in the detection rates for compromised cards, reducing risk and fraud-related costs for the company.
“This is quite remarkable because it does translate into hundreds of millions of dollars of risk avoidance,” he says. “They started with a risk-based problem, and then they used a competitive advantage that they’ve established already to solve it.”
This case, as well as the US banking case, exemplifies how having high quality data and clear business solutions supported by technology – rather than a technology-first solution – leads to higher quality outcomes, Ennamli says.
“They start with clear problems, and the solution is a business solution supported by technology,” he says. “That's quite important because if you start with a technology solution, it will become obsolete at some point.”
Success requires a hybrid approach
He adds: “The idea is to understand all different solutions and navigate between them, combine them in a way that is the right one for the problem you have, for the data you have and proportional to the human oversight that you can apply.”
And human oversight is key, he says, to successful AI-powered fraud and financial crime prevention.
“Very few people know what's under the hood in any AI solution. So I think it's a very critical must-have control to keep human oversight,” he says. “Whenever there is a critical decision that is made, AI is here to complement humans, not to replace them.”