Among other outcomes, 20% fewer of its customers experience false positives
An unfortunate reality is that many bank customers have experienced some sort of financial fraud at some point in their lives, and many are concerned about it happening again. Further, fraud is costing banks billions per year.
As the prevalence and sophistication of fraud continues to rise, banks are challenged with keeping pace. In fact, fighting financial crime is consistently a top trend and/or priority cited by bank executives in CGI’s annual Voice of Our Clients research.
Artificial intelligence and machine learning are helping banks to win this important fight. One example is a major Canadian bank that CGI has worked with for the past 15 plus years. We helped the bank enhance its rule-based fraud detection by adding machine learning, and the business outcomes have been significant.
Rule-based fraud detection limitations
The bank’s rule-based fraud detection approach had been in place for some time, but it had limitations. It was a challenge, for example, to develop rules, especially as fraud types and techniques rapidly evolved. Further, subsequently modifying rules was time consuming and costly.
As a result, the bank had to deal with false positives and their negative impact on customers. For example, legitimate credit card transactions would be improperly flagged as fraud. False negatives also were an issue, resulting in the bank’s mistaken payment of fraudulent transactions.
Rules that helped to reduce false positives often increased false negatives and vice versa. So, the bank was forced to choose between providing the best possible customer experience and protecting itself from financial loss.
Batch versus real (or near) time processing
Batch processing was another hindrance to the bank’s fraud detection efforts. Transactions were reviewed in batch overnight, enabling fraudulent ones to pass through prior to detection. Real or near time processing wasn’t possible because the bank had disparate systems generating disparate data that had to be synthesized before it could be analyzed. This took time, preventing the fast detection of fraud.
Overcoming detection hurdles with machine learning
To help the bank overcome these challenges, CGI first looked at its data capabilities, implementing advanced data engineering techniques to eliminate batch processing and enable the near real time processing of data. At the same time, we improved the quality of the bank’s data.
Next, we introduced supervised, semi-supervised and unsupervised machine learning models to supplement the bank’s rule-based engine and eliminate its sole reliance on rules. By combining machine learning with rules, the bank could implement a “challenger” and “champion” method to examine the results of each and determine which works best for specific transaction categories. Further, with machine learning, the bank could continuously introduce and test new machine learning models in response to new fraud types—an easier and less costly effort than changing rules.
Outcomes generated from our work
CGI’s partnership approach helped to build trust with the bank. We were able to work with the bank’s current systems (versus replacing them) and showed the bank ways to maximize the systems’ capabilities. Further, we modernized the systems by adding new data engineering and machine learning technology. As a result, the bank was able to detect fraud more accurately and faster. In fact, the number of bank customers experiencing false positives declined by 20%, improving the customer experience. At the same time, our enhancements increased the efficiency, reliability, and security of the bank’s systems, driving operational excellence, increasing the capacity of branch, contact center, and investigative teams to fight fraud, and saving costs.
We also worked closely with the bank’s teams to train them on machine learning algorithms, tool investments, and ways of collaborating with other teams and data scientists on machine learning model implementation and testing.
The success of our machine learning work led to new work with the bank in other areas, and CGI, to this day, continues to deliver a range of services and solutions as the bank’s long-term partner.
