AI Can Deliver Many Benefits in Financial Services but Also Comes with Risks

Financial services organizations are showing keen interest in artificial intelligence. According to a recent report by the Economist Intelligence Unit (EIU), 85 percent of banks have a “clear strategy” for incorporating AI into their products and services. Almost half (46 percent) of bank executives said these initiatives can help them achieve their objectives “to a great extent.”

AI’s potential cuts across many aspects of financial services operations. At the most basic level, AI can help financial services organizations streamline and automate traditional processes to achieve new levels of efficiency. That’s only the beginning. AI can also enable the development of innovative products and services. It allows financial services to tap into vast amounts of transactional and unstructured data to create a 360-degree view of customers and provide more personalized services.

That has become a necessity in today’s hyper-competitive market. Banks can no longer rely on longevity and asset size to attract customers. Today’s consumers expect a personalized experience, and those organizations that can deliver will gain competitive advantages.

Interested In Learning More?

Improving Fraud Detection

Fraud detection and prevention is a primary use case for AI in financial services. The EIU study found that 57.6 percent of banks use AI for fraud detection, and another 7.8 percent plan to adopt it within three years. A 2021 KPMG study found that 93 percent of financial services leaders are confident that AI can detect fraud more effectively than legacy solutions.

Traditionally, banks relied on systems that screen names and monitor transactions according to Anti-Money Laundering rules. These rules-based systems generate a large number of false positives. According to J.P. Morgan, false positives account for 19 percent of fraud costs due to lost revenue, customer dissatisfaction and the effort required to review transactions manually.

AI-enabled tools are not only more accurate but enable a more proactive approach. AI can identify suspicious relationships and anomalies in transaction patterns, enabling financial services organizations to prevent fraud rather than detect it after the fact. This is rapidly becoming essential given the escalation in fraud-related crimes.

The Problem of Bias

Financial services organizations that adopt AI should also be aware of the risks. Algorithmic bias is a major concern. AI systems that are trained using biased data can make discriminatory lending decisions and perpetuate existing biases in credit reporting and scoring. Bias can also result from the biases of the AI developers. In 2019, for example, Goldman Sachs came under fire when its AI-enabled tool approved fewer credit lines for women than men.

This can present serious consequences for organizations in the heavily regulated financial services industry. These organizations should incorporate bias detection in their risk management processes. They should also be prepared to explain decision-making processes to consumers and regulators.

There are also ethical considerations. Decisions about investment opportunities or creditworthiness have a significant impact on people’s lives. Who should take responsibility if an AI tool makes a serious mistake?

Security and Privacy Risks

The adoption of AI brings privacy and security risks to financial services firms. Data used in AI tools must be anonymized, but the tool could identify individuals through inferences. Data leakage is also a significant concern. Attackers can extract data from the training dataset even with read-only access. This could include sensitive financial information.

AI comes with the traditional cybersecurity issues of digital systems as well as novel threats. For example, threat actors could manipulate the AI system to extract data or make bad decisions. In a data poisoning attack, malicious data is added to the training dataset, causing the AI model to learn to recognize or classify information incorrectly. In an input attack, data is manipulated to mislead the AI system during operation.

Regulators are justifiably concerned about these and other threats. If your financial services organization is looking to implement AI, DeSeMa can help you assess and mitigate potential risks. Let us help you develop a strategy for detecting bias and securing AI models and data.