Financial crime may be getting more sophisticated, but so are fraud detection systems. Whereas early versions, referred to in this article as first generation, rely on the risk department manually applying a set of rules to detect fraudulent activity, the next generation leverage artificial intelligence (AI) to make the process faster, more efficient and scalable. In a recently published report, Juniper Research predicted that global spend on next generation systems will surpass $10 billion by 2027 and generate cost savings of $10.4 billion.

This blog post outlines the evolution of fraud detection systems to help financial institutions (FIs) make an informed decision about which one is most suitable, both in the short term and the future.  

1^st generation fraud detection systems: rule-based

1st generation systems use rules to screen transactions for fraud. These rules are based on attributes, such as a purchase from an unusual location, an elevated frequency of transactions or transactions involving large amounts with unfamiliar or newly created accounts. The risk department creates these rules and manually applies them to historical data to detect anomalies and trigger alerts. By design, rule-based systems are linear and reactive.

1st generation systems face several challenges:

• The system is too rigid. Once rules are enshrined in a risk policy, the employees tasked with identifying fraud only look for activities that breach them. That allows criminals to stay one step ahead of FIs, so 1^st generation systems must be constantly updated manually to detect new patterns
• They tend to block a high volume of legitimate transactions suspected of fraud, known as false positives. The risk department must investigate these alerts before they can be classified as such and processed, much to the customer’s irritation. Meanwhile, suspicious or genuinely illegal transactions go undetected, which may prove problematic for an FI both in terms of regulatory fines and reputational damage
• 1^st generation systems rely on manual processes which are time-consuming, costly and potentially error-prone. For instance, following paper trails may require trawling through multiple Excel spreadsheets saved on different computers. These systems don’t scale well- risk departments must respond to spikes in fraudulent activity by adding to their workforce, putting additional pressure on human resources and training teams

Next generation fraud detection systems: siloed AI

Next generation systems combine a rule-based approach to detecting fraud with a risk-based approach (RBA), which involves assessing the risks faced by an FI and then setting controls according to their severity. The Financial Action Task Force (FATF) considers an RBA as key for implementing the recommendations it issued in 2012, and it regularly publishes sector-specific guidance on its website.

The risk department configures an RBA based on the FI’s risk appetite. It assigns risk scores to customers, which are combined and used to enhance controls as part of the enhanced due diligence process, where FIs must conduct a more thorough investigation into certain clients, such as those classified as high risk or high net worth. 

Next generation systems use AI models to analyse data, typically gathered from an FI’s own transactions. These models detect suspicious patterns that may indicate fraudulent activity and assign a score automatically, reducing the time required by the risk department to investigate a flagged transaction.

Juniper Research’s report, referenced above, highlights three ways AI enhances fraud detection systems:

• Next generation solutions are better equipped to scale because the only constraint on AI is computing power (readily available through the cloud). Conversely, humans are limited in what they can achieve due to time.
• AI accelerates the process of detecting fraud, which improves the customer experience and allows FIs to adapt to new payment types requiring real-time monitoring

• AI models are more efficient than humans at identifying patterns that indicate fraudulent activity and acting on it

However,  next generation systems also face challenges:

• They’re only suitable for FIs with big enough datasets to train the AI model
• The provider must modify the model occasionally to make sure it aligns with the FI’s risk policies and to optimise it for speed and performance
• Next generation systems are costly and slow to deploy because they’re developed on a case-by-case basis using siloed datasets 

Further evolution of fraud detection systems - centralised AI (collective intelligence)

Centralised AI systems also leverage AI to detect fraud, but they train their models with a centralised dataset, rather than a siloed one. This dataset consists of billions of data points gathered from a variety of sources including payment providers, merchants, issuers, acquirers, marketplaces and third parties.

The richness of this data- it isn’t just diverse, it’s updated in real-time- produces powerful network effects which allow FIs to benefit from each other’s efforts to detect fraud. These network effects mean centralised AI systems can identify suspicious activity and unusual transaction behaviour in a matter of seconds and produce results 30 times more accurate than traditional systems.

There are several benefits to using a centralised database with AI and machine learning for fraud detection and prevention. Some of the key benefits include:

• Increased accuracy: By using a centralised database, the AI and machine learning algorithms can access and analyze a larger and more diverse dataset, which can improve their accuracy in detecting fraudulent activity.
• Improved efficiency: A centralised database can automate many of the manual processes involved in fraud detection and prevention, which can save time and increase efficiency.

To learn more about Fraudio’s fraud detection solution, set up a call to request a Proof-of-Results based on a sample of your FI’s historical data.