Shaping our approach to ethical, safe and responsible AI
September 2024
How to improve the effectiveness of fraud management
It’s no secret that financial crime, fraud and scams are not only a significant current problem for both firms and consumers, but one that is due to grow rapidly in the coming years. Fraud represented 40% of all crime committed in 2023 with an estimated £6.8 billion cost to society in England and Wales.
This situation is only being made worse due to the cost-of-living crisis, with consumers more willing to take risks, whether that be making ‘too good to be true’ purchases online, accepting opportunities that offer quick payouts, or romance scams resulting from individuals feeling alone and isolated.
Mix this with the rapid advancement and accessibility of technology, in particular Generative AI, and scams are becoming more realistic and security layers becoming easier to breach when in the hands of bad actors. For example, it may be exciting and fun to create fake images of yourself with interesting backgrounds using AI, but in eyes of the fraudsters, that’s fake IDs, synthetic documents and an opportunity to create fake bank accounts.
Whilst all of this presents significant challenges for financial institutions, it is more important than ever to give consumers the best user experience possible with a smooth customer journey. So, how do we strike the best balance between these two drivers?
The adoption of Artificial Intelligence (AI) and Machine Learning (ML) within fraud detection and prevention systems has meant large improvements in the battle against fraud. The use of anomaly detection algorithms or classification models that provide banks with a probability score are prevalent within firms, providing them with a tool to reduce the number of fraudulent transactions that occur.
AI solutions can identify underlying trends that are simply too difficult for a human to see, mainly due to the vast quantities of data available to fraud prevention analysts, along with the imbalance of fraudulent versus genuine transactions.
Although AI solutions are key enablers in the fight against fraud, they are not adaptable or transparent, making it difficult to react to emerging fraud threats or understand why a high-risk score has been produced. This is why rulesets are still fundamental to the operation of fraud prevention.
Rulesets give the organisation ultimate control over their risk appetite, alert volume and end user experience. They are easily interpreted, meaning they are easier to audit, but they can also be adjusted quickly when required.
The question is, how do we make the most of both technologies?
By combining the two approaches discussed, organisations can create a hybrid approach to combatting fraud and scams. By creating rulesets that factor in AI and ML based risk scores, a solution that is both transparent, adaptable and accurate can be implemented. This methodology is already being used across different organisations and is seen to be incredibly effective when compared to either technique used in isolation.
However, there is a problem with this methodology. Rulesets combined with AI become very complex, with millions of possible combinations and permutations of rulesets available. In order to adjust rulesets to changes in the fraud environment or the business drivers, thresholds need to be changed. Data Science communities are already addressing this problem with methods such as grid searches, Bayesian Optimisation or even more advanced techniques such as Reinforcement Learning.
This then presents a further issue. Rulesets get tuned, adjusted and updated over time, moving them further away from the very reason they were created. Organisations then find themselves updating legacy rulesets that are out of date, ineffective and far too complex to manage. So yes, we can continue to tune the ruleset, but we are better off redesigning the ruleset for the intended purpose at that moment in time.
By using AI to automate the design and construction of these rulesets, the benefits of hybrid rules are maintained, but with the added value of complete confidence that the rule design cannot be improved. An added benefit is that due to frequent and complete rebuild of the designs, they are no longer static, with gaps in the detection layer that fraudsters can expose.
This new technology provides a complete end to end optimised fraud management process that allows financial services organisations to stay in control, whilst having complete trust that they are providing the very best service for their customers.
Find out how this technology can help you achieve your fraud goals.
Originally posted here
