To overcome increased risks from financial crime, fraud expert teams have long been working with AI-powered solutions. Without such tools, inspecting all financial transactions for possible criminal activity would not be possible. But not all solutions are equally effective. The future of fraud prevention lies in intelligently combined multichannel monitoring and Hybrid AI, i.e., the optimal combination of data, sophisticated rules, and competent users.

The term Artificial Intelligence (AI) has always been a source of confusion and controversy. Unfortunately, there is no mainstream to guide the discussion. The so-called “General Artificial Intelligence” is the most prominent type of AI that attracts users’ attention and (in the most catastrophic scenario) is projected to replace humans in various functions, especially in the workplace. The goal of this is to create a robot or android that assimilates, talks and even reacts like humans. In this sense, AI-enhanced assistants, such as Siri, Alexa or Cortana, are first examples of this approach.

Currently, most AI programs offer us narrow-minded “special solutions” that can beat humans at chess or can master some discrete tasks to solve specific business problems. This practical type of AI uses machine learning techniques and brings some value to the different industries where it operates. With ChatGPT and other trained language models, technologies have emerged that seem to push these boundaries to new levels with many use cases of this technology only on the horizon. Nevertheless, effective fraud prevention needs more than just machine learning, which most often is used as a synonym for AI.

Understanding Hybrid AI

But what is machine learning (ML)? In simple terms, ML comprises a set of tools that use algorithms to learn from and adapt to data, allowing computers to find hidden insights without being told where to look. Classic ML is dependent on the maturity of the ML models they use. These must be intensively trained before they can be used for the first time with a reasonable hit rate. The time required is not available to financial institutions in the age of instant payments. If a bank’s fraud prevention team, for example, relied only on ML algorithms to learn fraud patterns, the criminals would have plenty of time to cause a lot of damage.

Currently, specialized software based on Hybrid Artificial Intelligence, combining the benefits of data-driven AI (e.g., ML) with knowledge-based AI (e.g., Fuzzy Logic or dynamic profiling), is used to effectively identify the risk and fraud certain organizations are executing in the financial market. In this sense, machine learning models have proven to be a powerful tool against this scourge. However, they require a great deal of information and experts behind them to be used to their full potential.

A hybrid approach, on the other hand, allows complementing machine learning models with expert knowledge, achieving immediate and more reliable results.

Hybrid AI in real-time fraud prevention

Current systems that maintain dynamic profiles for different entities (e.g., a bank account) can detect “potentially fraudulent” transactions in real-time through advanced rules. Additionally, through rules based on fuzzy logic, independent rules can be created to help avoid the risk of a particular event. Thus, by being able to handle independent rules for different data within the same transaction, it is possible to create “fingerprints” for customers, identifying unknown patterns. Being isolated cases, these patterns could remain under the radar of a model based entirely on machine learning. As an example, imagine a transaction where the amount is 30% higher than the average amount drawn in recent months. There are also several new accounts involved for the customer, amounts close to the maximum daily amount allowed, unknown IPs, countries of the transaction different from the customer’s, etc.

Each of these transaction anomalies alone may be strange and suspicious, but not enough to raise the proper alarms. With the Hybrid AI model, advanced techniques and machine learning work hand in hand, making software using this technology a substantially more effective alternative for anticipating these types of problems.

A critical factor in making this approach effective is the immediacy with which these models can react. Few tools have the ability to work in real-time and provide a “millisecond” response to identify and prevent fraud before it happens. Those models combining all available technologies and techniques have a substantial comparative advantage over models focusing on only a few technologies. This allows organizations that use them to achieve significant savings and have a healthier and more satisfied customer base.