Battling cybercrime with AI to save millions of euros

The challenge

Cybercrime is rapidly on the rise. Both governments and businesses increasingly face pressure from criminals exploiting the Internet’s speed, convenience, and anonymity. In 2020, Cybersecurity Ventures projected global cybercrime costs to grow by 15 percent per year over the next five years, reaching USD 10.5 trillion annually by 2025 (link).

For several years, the operator searched for a way to document patterns of fraudulent behavior in purchases of its online products which cost them millions in euros each year. They knew that they were being attacked by organized cybercrime as they had previously tracked suspicious transactions and credit card numbers manually. The challenge lay in finding a link between these single transactions to prove a pattern in online purchases from various suspicious purchasers that acted similarly. Without being able to prove links between the fraud cases, the financial costs could not be covered under the operator’s insurance policy.

Under the operator’s insurance policy, only fraud cases above €100, 000 were covered. As its products are split into multiple small transactions ranging from €5 to €600, the fraudulent activity fell far below the threshold. The operator needed to show that the customer profiles that committed the fraud were connected in order to prove that the total monetary amount was far greater.

Together, the client and 2021.AI worked to:

• Prove collusion for fraudulent transactions
• Apply AI models to identify patterns in the data
• Report the model results with what the business already knew about the situation
• Present a scientifically-backed claim linking fraudulent transactions to the insurance company

The solution

They were able to make a breakthrough by applying AI to analyze thousands of transactions, numbers, dates, and more, and prove the links between transaction patterns and fraudulent behavior. A series of AI models were built to search for hidden patterns in each transaction. As the operator already knew which transactions were fraudulent based on information from financial institutions, a classification model was built to show the significant differences between the fraudulent transactions and those that were non-fraudulent. A clustering model was then developed to define certain groups of behavior in fraudulent transactions, and identify patterns in certain actions that could be grouped together. The final model was trained to separate out fraudulent transactions and then identify multiple patterns in them which could be linked to the different clusters.

The results

With the ability to separate fraudulent transactions and identify the patterns, the organization proved the links between fraudulent behavior and specific profiles, saving the operator money and increasing its revenue and sales. Thanks to the models, the monetary numbers shown to the insurance company were above the threshold. The model results were presented in a comprehensive report which displayed the results, the assumptions, and applied methods, as well as the interpretation of the results, including inside knowledge from the internal fraud management team.

In the end, the client achieved their goal, successfully presenting the case to the insurance company, which resulted in coverage amounting to several million euros. The operator can now continue to spot and link fraudulent behavior in the future, which would be virtually impossible to do manually.

Project highlights

• A public transport operator used machine learning to support arguments for insurance claims
• 2021.AI collaborated with the operator’s fraud management team to create a series of models that linked fraudulent transactions
• The two organizations worked together to successfully recover several million in euros

About the company

The transport operator provides passenger transport services and has a long tradition within rail transport, operating railway services since its foundation in 1885. They provide long-distance and regional train services, as well as public transportation.