Multinational technology corporation IBM calculated that 72% of business leaders cited fraud as a growing concern in the last year, that $44 billion will be lost worldwide due to fraud by 2024, and that a quarter of e-commerce sales transactions that were declined by artificial intelligence (AI) were false positives. AI has become the leading tool for fighting fraud, but it can still be improved upon.
Image Credit: TippaPatt/Shutterstock.com
AI in the Fight Against Fraud
In the past, rule-based engines and simple predictive models were used to computationally identify the majority of fraud attempts. But these methods have not kept up with the increasingly sophisticated nature of fraud attacks today.
With a proliferation of digital technologies at criminals’ disposal, fraud has grown in both scale and severity over the last few decades. Large criminal organizations and even state-sponsored groups use AI-like machine learning (ML) algorithms to defraud digital businesses for millions of dollars each year.
The technology that has enabled this increase in scale and severity of fraud attacks is also vital in the fight against fraud. AI and ML platforms can combine learning from both supervised and unsupervised machines to provide a weighted score for any digital transaction in milliseconds.
Years and even decades of transaction data can be analyzed by AI-based fraud prevention systems in around 250 milliseconds. This analysis is quickly employed to output risk scores calculated with AI.
AI also takes emerging activity into account. Before AI, fraud prevention systems relied on only past activities to analyze the risk of fraud. But AI is capable of taking emerging activities, trends, and behaviors into account when analyzing transactions.
Real-time Fraud Detection
These benefits of AI in the fight against fraud combine to allow for real-time fraud detection, saving consumers and digital businesses millions of dollars.
Before this capacity was introduced to the retail sector, fraud would generally only be detected once chargebacks were issued up to two months after the fraud took place.
This real-time detection helps analysts set threshold scores to ensure the best possible balance between low false-positive rates, which negatively impact genuine sales, and losses from fraudulent activity.
Nuanced Fraud Attacks
As well as adding real-time detection capacity to fraud analysis and prevention, AI can also spot and thwart more complex, sophisticated, or nuanced fraud attacks.
Attacks abusing refer-a-friend features, promotions, or seller confusion in the marketplace – all very difficult to identify as they are so similar to normal activity – can be reliably spotted by AI fraud prevention systems.
Without AI to more accurately detect nuanced fraudulent activity, digital businesses working at scale are forced to increase the threshold limits for rules-based fraud prevention, which inevitably results in more false positives and lost sales.
The greater accuracy provided by AI ensures this threshold limit can be set more appropriately without letting fraudulent activity through the net.
AI Fraud Prevention is Necessary for the Massive Scales at Work in the Digital Economy
Massive scales of operation are made possible by digitization – as transactions, products, and services can all be replicated at no or minimal extra cost in digital environments.
As digital businesses – selling goods or services – rely more on AI for fraud prevention, they will gain more control over chargeback rates, declined transactions, and losses from fraud. This enhanced ability to analyze and control complex, dynamic aspects of digital businesses is necessary for companies growing to massive scales.
AI-based fraud prevention is also instantly customizable. Users can adjust parameters to ensure specific and changing business objectives are met. For example, periods of peak fraud activity such as online sales can be guarded more carefully than periods of low fraud activity, or particularly vulnerable aspects of digital business operations can be guarded against more vigilantly.
Modern industries like gaming also have different requirements and tolerances for fraud detection. Rapid transactions are necessary in some gaming applications, and slow fraud detection will lead to a poor experience for customers.
Low-margin, high output digital businesses such as e-commerce stores and most dropship retailers benefit from AI-based fraud analysis to ensure tight margins do not exceed profitability with chargeback rates and fraud losses.
In all of these examples, friction – the boogeyman of the digital economy – is reduced, and consumers have a more satisfying transaction experience.
Trends in AI-based Fraud Detection and Prevention
NVIDIA has been providing full-stack accelerated computing for AI and other data-consuming high-performance computing applications to the financial industry for nearly two decades.
The company’s AI-based products help the largest banks, credit card suppliers, and insurers around the world to increase revenue, reduce costs, and mitigate risks.
Kevin Levitt, a director at NVIDIA, cites three key trends for AI-based fraud prevention in a 2021 interview with the Fintech Times.
Levitt says that increasingly sophisticated AI models will continue to replace the basic rules based engines of classical computational fraud prevention. Deep learning can work with more data and provide more accurate outputs, including and especially in fraud prevention.
The second key trend Levitt cites is a move by banks away from structured, tabular data into more unstructured data such as images and voices in their fraud prevention systems. Intelligent Voice, for example, analyzes callers’ voices to flag identity fraud.
Finally, Levitt says that new technologies powered by GPUs (graphics processing units) will help to protect consumer privacy while allowing data to be shared for fraud analysis. Financial institutions will be able to share encrypted data privately to give more power (in the form of more data) to AI-based fraud prevention systems.
Continue reading: AI-Enhanced Innovation.
References and Further Reading
IBM (2022) AI improves fraud detection, prediction and prevention. [online] Available at: https://www.ibm.com/uk-en/analytics/fraud-prediction.
Columbus, L. (2019). Top 9 Ways Artificial Intelligence Prevents Fraud. Forbes. Available at: https://www.forbes.com/sites/louiscolumbus/2019/07/09/top-9-ways-artificial-intelligence-prevents-fraud/?sh=4b19f0d714b4.
Harrison, P. J. (2021). How Can Artificial Intelligence Fight Fraud with NVIDIA. The Fintech Times. Available at: https://thefintechtimes.com/how-can-artificial-intelligence-fight-fraud-with-nvidia/.