Fraud has reached the highest levels on record, affecting more organizations than ever. The scale of the problem was revealed in last year’s PWC Global Economic Crime and Fraud Survey. Where 49 percent of the 7,228 businesses across 123 territories, contacted by PwC, reported that they had experienced fraud and economic crime, over a two-year period.
Payment Fraud is also on the rise, alongside our awareness of it. In fraud professional circles, it is now recognised that adoption of a data-driven approach, which includes A.I, is required to effectively tackle fraud. A.I is now seen as the next evolutionary step for a data-driven approach, following from human data analysis, and I fully expect that blend of human input and A.I to deliver the performance required to counter the evolving payment and fraud landscape in the future.
Many organisations currently utilise A.I to augment, not replace, their human interactions. Viewing A.I as a productivity and scalability enabler. A.I is seen as an additional member of the ‘team’, which can operate at a similar level to the human equivalent and operates 24×7. As experience and confidence in A.I grows, organisations are becoming more willing to place increase emphasis on A.I to fulfil specific areas of the fraud management lifecycle.
It’s clear that certain elements of the fraud lifecycle will continue to be heavily supported by A.I in the future, for example, the automation of repetitive activity, advanced pattern detection and scalability of completing tasks in parallel. However, there are other aspects of fraud prevention, such as investigation and business strategy, that require ideation and innovation, and will remain best tackled by humans.
How A.I is helping to tackle fraud today
Traditional, manual, rules-based fraud detection systems require full time manual effort to keep up with changing fraud patterns. A.I supplements these systems, by keeping up with trends as they change, by analysing the data and understanding any underlying patterns, which experienced fraud analysts are unable to do due to the large volume.
A.I models remove the need to continually and manually manage a fraud system, by adding and removing rules, analysing data and other manual activities. Providing operational benefit on top of fraud risk benefit. By incorporating A.I tools into the fraud detection ecosystem, businesses experience efficiency gains throughout the entire risk management sector; as time costly, manual processes become automated and highly skilled individuals can be freed up to work on other revenue generating activities.
A.I is an excellent tool for automating the day to day upkeep of a risk system. Fraudsters are clever and are constantly developing new threats. A.I is the best defence against these fast-changing trends, with models constantly learning, and adapting to better understand cardholder spending, making it better at detecting anomalous activity. Models can adapt to new threats and traditionally ‘difficult to detect’ types of fraud, due to the statistical nature of A.I, which can uncover those patterns. A.I models, and processes, should be applied in big data environments, where it is expensive, not feasible, or impossible to manually analyse the data for fraud trend changes.
New data streams are also particularly well suited to application of A.I, since, in many cases, little is known about fraud by the fraud manager. A.I can extract fraud patterns in the data and use that intelligence to prevent further threats, while learning new fraud patterns, as they happen. This is done much faster than any manual process. Enabling a faster, safer ‘product to market’ process.
A.I is best utilised for monitoring activity on a temporal basis, where there is either historical or ongoing activity that needs to be assessed to identify a change in behaviour. Static fraud detection, such as application fraud, that has been traditionally tackled by scorecard approaches can also be tackled using A.I.
However, drilling down further, the capability to detect different types of fraudulent activity is less pronounced with AI. My team and I have conducted experiments, in conjunction with our clients, to identify whether A.I is better suited to certain types of card fraud – lost/stolen cards, skimming or collusion etc. We found that there was no significant difference in detection capability across different fraud types.
While traditionally A.I was seen to require a fully employed data scientist resource to develop, deploy and maintain AI solutions, that took months to realise benefit; there has been a paradigm shift towards self-service and increased speed of deployment. This has meant that business users (citizen data scientists) can manage A.I technologies and deploy solutions, within a matter of hours. This paradigm shift has resulted in a wave of interest and adoption by business, who were previously unable to commit resources towards the technology.
The future hold for its use in tackling fraud?
The use of A.I for fraud detection and prevention will become increasingly prevalent, as fraud managers look to increase efficiency and fraud detection rates; while craving out time to pursue other opportunities and business needs. A.I systems are continually improving on their performance and will adapt to add in other business-add features, such as marketing opportunity dashboards.
Over time, A.I systems will be increasingly relied upon, as they become more trusted to carry out effective fraud detection. This coupled with data, that is getting bigger, will be increasingly difficult to manually analyse for the same level of fraud performance, and we will see A.I utilised to a higher degree. Once A.I technology has developed to the point where the system performance can no longer be improved with manual input, A.I will become a necessity to maintain adequate system performance and business goals.
In summary, A.I will become more central in tackling fraud, as institutions struggle to scale their operations, while keeping pace with increasing volumes of data, transactions and channels. Recent data from UK Finance reported that payments in the UK in 2017 totalled 38.8 billion, split amongst nine main payment types. This is predicted to rise 5% to 40.9 billion by 2027, during which cash usage, as a percentage of all payments, will reduce from 34% to 16%. This rapid increase, combined with initiatives that speed up customer interaction (expedited provision of goods, faster payments etc.) will mean that there are more fraud assessment decisions that need to be made in real-time. The only way to achieve this, and maintain profitably, is to use A.I to support the human workforce.
To read the original article;