AI is proving its worth in many cybersecurity use cases. For example, it can help detect malware, monitor traffic for anomalies and enable human analysts to prioritize security alerts — all of which can reduce costs and human error.
“AI has been adopted by many different companies, industries and countries,” said Meng Liu, analyst at Forrester Research. “Adopting AI-based models is quite relevant to your competitive advantage.”
Now, AI has its sights set on improving the fraud detection and management landscape — an area fraught with difficulty as the number of online and mobile transactions surges.
Organizations today need to implement processes that detect fraud quickly and accurately in real time. AI is the answer, Liu and co-authors Andras Cser and Danny Mu explained in their report, “Artificial Intelligence Is Transforming Fraud Management.” In the report, the analysts map AI subsector technologies to their use cases for fraud management and offer relevant case studies.
The benefits can be aplenty, the analysts reported, but organizations deploying AI for fraud management must also be aware of the potential challenges.
AI use cases in fraud management
Several use cases for AI in fraud detection and management were discussed in the report. For example, AI can improve the accuracy of transaction monitoring. The analysts described how financial services provider FIS worked with Brighterion, an AI company owned by Mastercard, to improve its anti-money laundering capabilities. The provider now uses AI to vet risk when onboarding new vendors, for example.
In other use cases, AI can improve the efficiency of fraud investigations by streamlining and prioritizing alerts. AI can also be used in biometrics to authenticate users before a transaction can even occur, lowering the chances of malicious transactions.
Different AI fraud management use cases require different capabilities, the analysts noted — an important consideration when implementing AI for fraud detection and management. For example, e-commerce transaction monitoring requires faster response times and more precision than transaction reporting. Other capabilities to keep in mind include training data availability and accuracy, as well as adopting processes to ensure AI models improve over time.
The benefits of AI for fraud management
Implementing AI for fraud management comes with two major benefits: Fraud is more quickly detected, and AI models get more accurate with time.
The number of mobile and online transactions has grown steadily over the past two decades and significantly since the start of the COVID-19 pandemic, Liu said. “These transactions are happening in real time, which means fraudulent activity also becomes real time.”
Historically, fraudulent purchases were discovered by human analysts after transactions were processed. AI can improve this by detecting and preventing fraudulent activity as it occurs and then blocking malicious users and transactions. For example, if a cybercriminal tries to use stolen credit card information to make an online purchase, AI will detect the fraudulent order and stop the transaction from processing. AI-enabled fraud detection also shrinks organizations’ expenses by reducing the amount of money spent on reimbursing stolen funds.
Beyond detection speed, AI models also improve their own accuracy as they mature. Unlike traditional rules-based models, which can create an influx of false alerts in an industry already overwhelmed and understaffed, AI models become more reliable with time.
“The more data you put into an AI model, the more intelligent it can become,” Liu said.
Data improves AI efficiency by recognizing patterns and analyzing connections between data sets. This accuracy also helps organizations decrease their response times to fraud alerts.
AI challenges abound, principles to keep in mind
While AI can be a game changer for fraud management, it is still prone to one of AI’s greatest challenges: bias. Skewed data and subjective engineers can undermine an AI model’s ability to detect fraud.
To prevent this, Liu said, organizations should keep three principles in mind when implementing AI into their fraud detection and management processes:
- Objective models. AI models need to be objective and fair. Biased models lead to inaccuracies, which makes the entire model redundant. Liu suggested creating a hybrid model. “Having both an in-house dedicated team, plus an external vendor, would be the best way for an organization to start building its AI-based fraud management capabilities,” he said. Internal members understand the company’s needs, such as industry and compliance regulations, while an external vendor provides a customized but well-tested model.
- Third-party sources. More data means less biased and stronger AI models over time. Liu recommended using open source data when training AI. Including data from a credit bureau, for example, will help augment models in the financial industry. For companies concerned about data privacy and security, federated learning is an option. It enables multiple organizations to share the same data training results without exchanging actual data, according to the report.
- Augment — don’t replace — analysts. While adequately trained AI models are effective at preventing fraudulent activity, human analysts should still view alerts and perform analysis to understand why content was flagged. By understanding alerts, analysts will have a clearer idea of how the AI model learns and works.
Outlook of AI in fraud management
The future of AI for fraud detection and management will be beneficial, but adoption rates will vary across companies, industries and countries, Liu said.
“The technology, retail, e-commerce and financial sectors will become the forefront of AI-based fraud management models,” Liu said. Digitally intensive industries will rely more heavily on AI fraud management, Liu added, which will enable them to develop models faster. In traditional industries, such as manufacturing, AI will develop more slowly due to the lack of digitized data and prevalence of offline interactions.
The development of AI for fraud detection and management will also vary by region. Emerging markets, particularly Southeast Asia and Africa, will likely continue to use human analysts for longer than more developed countries, Liu said. In the short term, he added, human analysts are more cost-effective in regions deploying AI more slowly.
Liu encouraged those dubious of AI to give it a chance. “Adopting AI models can give you a lot of advantage over your peers, so don’t be hesitant,” he said.