There’s no doubt that the continued development of Artificial Intelligence (AI) will radically alter the front and back office of financial institutions.
An area that will significantly benefit from AI is fraud detection. Right now, the high rate of false positives in banks indicates that current fraud detection methods are failing. In fact, it is estimated that only 1 in 5 transactions declared as fraud are truly fraudulent, and it is not uncommon to have false positive rates as high as 80%.
But whilst AI, and more specifically Machine Learning, is a promising new technique, it is constrained by the reliance on historical data in existing fraud detection systems; we can’t improve what we can’t measure.
The problem with historical data
Machine Learning, a subset of AI, involves algorithms that progressively improve themselves by using data; the more data they consume, the better they get at spotting patterns. However, restrictions occur when these algos are based on historical data, such as bias, structural changes and privacy concerns.
Bias: If algorithms are trained on historical data there will be inherent bias originating from historic trends or human involvement.
Structural changes: Algos trained on historical data may not be equipped to detect unseen, novel and evolving types of fraud or adapt to changes in regulations.
Privacy concerns: Under current privacy regulations like GDPR, firms are restricted from using even their own customer data. Where this kind of sensitive data is concerned, people must give their explicit consent to its processing – significantly hindering the use of Machine Learning.
If financial institutions continue to rely on historical transactional data for fraud analytics, they will soon bump up against these limitations of Machine Learning and AI.
Creating fraud training datasets with no historical precedent
There is an approach that has the capacity to address the limitations of historical data when developing fraud controls: agent-based simulation. With agent-based simulation, financial institutions can produce a virtual environment that resembles a bank payment system, using synthetic data based on real customer transactions with no historical precedent.
This synthetic data will contain no personal information or disclosure of legal or private customer transactions, so it is completely compliant with privacy regulations like GDPR. It has the added benefit of being easier to acquire, faster and at less cost for experimentation, even for those that have access to their own data.
Importantly, financial institutions can use the synthetic data to measure the number of innocent people wrongly tagged (False Positives) and increase the number of criminals flagged (True Positives), while knowing the levels of undetected fraud.
Unveil the hidden fraud to measure the real improvement
AI and Machine Learning can use objective functions to set goals for fraud control systems. But without visibility of the hidden fraud, there is no way to measure the real improvement of these systems.
By using simulation, firms can generate labelled instances of simulated fraud to test, benchmark and measure the real improvement in reducing the hidden fraud.
Understanding future fraud
Unlike linear models that are tied to past data, with agent-based simulation, firms can introduce Machine Learning agents into their fraud detection systems. These agents will be positively reinforced when they successfully detect fraudulent transactions, and over time will start to uncover fraud that hasn’t even been committed yet. This means that, for the first time, algos can be optimized to run in environments that infrequently occur, such as in rare fraud cases or events such as Brexit.
This is only the first step to set things in place for the approaching industry 4.0 revolution. Soon, agents will not only be used in simulated environments – they will be deployed into real world systems to analyse data in real time, and flag suspicious and fraudulent behavior to build strong cases for prosecution by authorities.
By understanding future fraud before committing valuable resources, banks and other financial institutions can use precisely the right approach to minimize financial fraud, as well as protecting clients from arbitrary, costly and ineffective fraud controls.