Financial institutions increasingly incorporate AI solutions into their workflow systems to manage risks in a sea of common frauds and scams. 

Machine learning (ML) models, which use AI and are trained on past data, can employ pattern recognition to automatically identify and prevent potentially fraudulent transactions from occurring. They may also require that human agents carry out additional authentication procedures to confirm the legitimacy of a suspicious transaction. 

Furthermore, AI technology can use predictive analytics to forecast the types of future transactions an individual may engage in and detect if a new transaction or behaviour deviates from the norm.

Evolving Risk Management Strategies

Digitisation across industries has grown rapidly. However, Venkat Srinivasan, chief analytics and risk officer at Bureau, told AIM that the methods for managing risk and compliance initially progressed slowly. However, particularly in the last decade, the focus on risk has significantly evolved over the past five to 10 years. Now, controls do not assess risk as a snapshot in time.

He highlighted that companies have shifted their focus towards compliance to combat fraud but have overlooked risk management. “AI allows you to identify this outlier much faster. In many ways, while the fraudsters usually are one step ahead, the modern control tools and behavioural tools are catching up very fast.” 

Srinivasan believes that fraudsters now know how to lie low, going into dormancy before quickly ramping up their activities. It’s similar to guerrilla warfare against the financial system. Fraud is increasingly treated as a service, with people engaging in it for profit.

Through various methods, AI fintech can help safeguard individuals against financial losses caused by multiple forms of fraud, such as phishing scams, identity theft, payment fraud, credit card fraud, and other types of banking fraud.

In 2022, digital wallets overtook credit cards to become the most popular payment option for US online consumers, making up 32% of e-commerce transactions, according to a Mastercard survey. A significant drawback of direct payments is that customers often bear the brunt of their fraud losses, although major banks may reimburse their customers to maintain strong relationships. Moreover, A2A transactions appeal to money launderers due to their frequently untraceable nature. Fraudsters can simply close their accounts after receiving the funds, the survey showed. 

Srinivasan said that today’s AI tools must quickly identify anomalies within data, and that it’s essential to look beyond individual cases and focus on the larger network of connections. Most fraud cases are increasingly organised, resembling syndicates, which requires understanding the individual involved and the network to which they belong. “For example, a single device may be used by multiple individuals, or a specific card or phone number might be reused frequently in a short span of time, creating a web of connections.”

Challenges and Innovations in Financial Security

The rise of e-commerce and online payment systems has made it challenging for retailers and businesses to discern a customer’s intentions or identify potential fraudulent transactions.

To tackle this, the Bureau has developed a system that allows multiple individuals to use a single device to identify these types of networks and their interactions. 

Srinivasan stressed that in this environment, fraud prevention measures must focus on these networks and consider how to manage both dormant periods and rapid fraudulent activity.

“It’s about finding these patterns at a very quick point in time and seeing the anomaly. And that is where AI has proved to be quite a boon. If anything, it is going to be the modern machine learning models and AI tools that are likely to catch these things right.”

According to him, what the company accomplishes at scale cannot be replicated by an individual, as they can process thousands of records in a very short period. Even with only 100 records and 100 people analysing them, it would still be challenging to find the “needle in the haystack”. This issue is not merely computational; identifying these networks is quite complex, he explained.

Srinivasan emphasised two key points. First, at a computational level, these models are highly effective and can generate information much faster than any human can. Second, even if the platform removed the computational aspect, there are patterns that humans simply cannot recognise, and that is where these models are deemed “superior”.

However, fragmented and manipulated identity data cause biases that could misclassify legitimate users as fraudulent. Incomplete data is a common issue that depends on how clients provide it. Engaging with clients is essential for understanding this data. 

To tackle these challenges, the Bureau compares new data against a “golden dataset” or benchmark, though this benchmark may also contain biases, making supervision necessary. They examine discrepancies by looking for patterns and potential skewness from off-peak seasons or targeted campaigns.

The Future of Financial Fraud Prevention

According to an IBM report, AI systems for banking fraud prevention are optimised for specific tasks. They are trained on large, curated datasets through supervised learning, which helps them identify patterns for particular functions. In contrast, unsupervised learning allows AI systems to draw insights from data without guided training. 

As per Srinivasan, while statistical techniques mitigate biases, understanding the data’s source is crucial. Although complete clarity isn’t always achieved, the goal is to identify data origins. The company also recommends that clients use their own datasets and benchmarks to uncover regional biases and encourage deeper analysis of skewness in applications from specific areas.

The risk officer emphasised that AI is likely to become the default mode of operation, driven by the vast amount of data and the complexity of emerging patterns. Distinguishing anomalies can be challenging, making AI essential for providing feedback during investigations.

The IBM report said that AI systems are especially beneficial for applications that need pattern recognition. Certain types of AI, referred to as graph neural networks (GNN), are specifically created to handle data that can be depicted as a graph, which is frequently found in the banking sector. GNNs can analyse vast amounts of records to uncover patterns across large datasets, allowing them to detect and prevent even the most intricate cases of fraud.

The rise of AI stems from its computational power and anomaly detection capabilities. While it is becoming the standard approach, safeguards are still necessary. However, human intervention will be important when explanations are needed, in buyer contexts, and for addressing ethical considerations, Srinivasan concluded.

The post What AI Sees in Financial Crime That We Don’t appeared first on Analytics India Magazine.