Fraud: In a real-time world can AI detect and prevent it in real-time?
While digitization in financial services is bringing many benefits, it is also making financial institutions more vulnerable. According to recent reports by McAfee, cybercrime and financial fraud is currently costing the global economy an estimated $600 billion per annum. In fact, the true cost could actually much higher as firms cannot accurately identify and measure losses due to fraud. The 2018 AFP Payments Fraud Survey, underwritten by J.P. Morgan, revealed that a record 78 percent of all organizations were hit by payments fraud last year. New initiatives such as faster payments increase the speed and impact of fraud. In fact, when the faster payments network was implemented in the United Kingdom, financial services institutions saw a 300% increase in fraud.
The fraud problem is global – for example India’s Banking Security and Fraud Cell reported that the total amount involved in bank fraud cases in the month of March 2018 was nearly double the amount reported for the entire year of 2017. The problem is clearly growing, however not all frauds are the result of external cyberattacks.
Insiders are involved in many cases. In the infamous Nirav Modi case, red flags – even those issued internally – were missed or deliberately ignored. The import and export transactions of the Brady House branch in the 12 months to March 2017 stood at $3.3 billion, 50 percent higher than recorded two years prior, largely because of the branch’s dealings with Nirav Modi and associates. An internal review report stated that the exceptional growth should have been noticed, but it wasn’t. The loss suffered was in the range of USD 1.8 billion which could have been about $800 million lower, had the fraud been discovered a year earlier.
The increase in the number and scale of these frauds – along with the delay in detecting them – is concerning banks and corporates alike. In a recent BAML survey, 65% of the banking respondents stated that they have higher payment security concerns compared to the previous year and 47% of corporate respondents mentioned that their most important AR/AP payments initiative are centered around fraud control.
Artificial Intelligence and Advanced Analytics Can Help
With greater accessibility of business data from internal and external sources, some financial institutions have started focusing on leveraging using advanced analytics in their fraud and fault prevention programs. In the past financial services firms typically focused on the quantification of financial impact when fraud or duplicate transactions were detected. However, advanced analytical tools – with built-in machine learning capabilities – can enable financial institutions to do much more. Machine learning systems can be trained using historical data patterns based on defined metrics. Any new incoming transaction can then be checked for the probability of fraud and if it significantly deviates from or breaks the past trends, it can be marked as fraudulent. As the data is accumulated, it aids the system in becoming more intelligent and effective. Duplicate transactions can be captured swiftly and potential instances of fraud can be intercepted and stopped before they happen, thereby increasing the effectiveness of fraud monitoring programs.
Machine learning- enabled systems present a host of advantage to banks, including:
- Higher Speeds – Traditional systems which use rules to decide whether to accept or reject a transaction, result in a time consuming and inflexible solution. They also increase the chances of false positives and damage customer experience. Eg: Blocking every payment above a certain threshold might appear to be prudent but it will annoy customers. Machine Learning based fraud detection solutions detect payments based on data patterns and can evaluate huge numbers of transactions virtually in real time.
- Enhanced Efficiency – Machines can handle various tasks in parallel and have generally been more effective than humans reviewing patterns to detect any fraudulent transactions. It also helps reduce false positives and helps identify new data patterns. They work 24×7 and can scale quickly to meet changing demand.
- Better Customer Relationships – Since financial security is a concern for most of the organizations, faster fraud remediation and tracking will help financial services firms to maintain better relationships with their customers.
- Regulatory Compliance – The rising regulatory costs and compliance breaches are a concern for the banks. While human scrutiny is valuable, AI can be deployed to perform lower level functions, and thus enable the organization members to invest time in more important decision making at higher levels.
Banks need to adopt a “Horses for Courses” Policy
While established banks already have gigabytes of customer data to get insights from, newly created banks do not have any historic data. Based on their particular requirements, an advanced analytics solution can help both sets of banks generate insights through supervised or unsupervised learning models.
In supervised learning, a set of algorithms are set to detect patterns in the existing historical data. The historical data cases can be labelled as fraudulent/not fraudulent and with repetitive information processing, the system can become more intelligent and predictive. Popular supervised learning models used in fraud detection are logistic regression, neural networks, and deep learning.
Unsupervised learning, on the other hand, can be implemented in cases where history of fraudulent transactions or labelled data is not available. Machine learning can be applied to identify patterns and create subsets of the data. This helps reduce the uncertainties over whether the fraud has been actually attempted or not. Data scientists apply unsupervised learning models to group data into clusters based on hidden correlations. Once the data is labelled, the idea is to apply this dataset to train the supervised models to detect fraudulent transactions. Express verification further helps by implying simple anomaly detection or by dividing the transactions into regular or suspicious ones. Regular can be processed further and suspicious can be then sent for advanced verification. Unsupervised learning models include anomaly detection using Gaussian Distribution, or Clustering.
Dealing with Real-time Challenges. In Real-time
Fraudsters have proven time and again that they will adapt their approaches as the situation evolves – from creating fake currency notes and fake credit cards to hacking into back end systems to raise card limits and issuing SWIFT payments. It is only a matter of time before they deploy AI in their approaches, indeed some commentators are referring to 2018 as “the year of the AI-powered cyberattack”. Against this backdrop it is imperative that sophisticated, self-learning, adaptive solutions are deployed. Advanced solutions with machine learning capabilities such as FinnAxia FarEdge help corporates and banks to not just capture duplicate transactions swiftly and effectively but also detect anomalies and potential instances of fraud before they happen Such solutions can bring a substantial difference in the way banks work and deliver customer experiences. As they become more intelligent as the volume of data grows , these solutions add more teeth to the battle with the fraudsters. And because they operate at machine speed, they can deal with the real-time challenge, in real-time.