Customs Risk Management Solutions

Home/Success Stories /

Classification

Computer Vision

Image Processing

Recommender Systems

Statistics

Highlights

Challenge

Create a comprehensive risk management solution that will swiftly detect fraud in international trade

Solution

5 In-house-built ML solutions that cover the full spectrum of risk management

About the Project

The Client was an international trade solution provider that wanted to enhance their rule-based customs risk management system with various ML solutions. They wanted to create a system that would help detect fraud in their transactions, create more efficient customs criteria, calculate risk scores for different entities and events, analyze the authenticity of documents, and more. The Portmind team combined a set of ML approaches, such as gradient boosting and image processing, to generate a comprehensive risk management product for efficient customs fraud detection.

CHALLENGE

Create a comprehensive risk management solution that will swiftly detect fraud in international trade

Fraud within the international trade process can take various forms. The most widespread types of fraud include misclassification (i.e., assigning goods to the wrong category), undervaluation (i.e., falsely declaring the invoiced price of the goods), and the falsification of documents.

Each country develops its own approaches and mechanisms to detect fraud. These include product valuation and classification validation, documentary check, and physical inspection of goods. In the traditional workflows, the risk specialists at customs manually check the declared information against what is stated in the trade documents. They confirm the validity of the submitted documents, physically inspect almost all the goods to make sure that the quantities, weights, and other descriptive information correspond to what was declared. Finally, they manually create risk rules and criteria to detect fraud.

Detecting fraud with these traditional customs procedures is expensive and time-consuming. Moreover, considering the limited number of available resources and the ever-growing number of transactions, it is becoming almost impossible to inspect all the transactions to identify all the fraud. This presents a challenge for the industry: how can we automate the process of identifying suspicious transactions?

SOLUTION

Using various ML approaches, the Portmind team developed an integrated risk management system that helps governments detect fraudulent activity in the international trade process, streamline risk management more effectively, and make data-driven decisions. To provide a comprehensive approach to managing risk in international trade, Portmind introduced the following tools to cover different aspects of fraud within the process. Entity Risk Profiling The ML solution makes efficient use of historical data to generate a risk score in real-time for each event or entity in trade, such as Consignee, HS code, Country of Origin, etc. The solution can be customized to create a profile for any entity, according to the Client’s needs. Based on the generated risk scores, the solution creates lists of high-risk or low-risk entities that are updated in real-time as it receives new data. The solution also provides a dashboard with statistical insights about the entity’s activity and displays anomalies. By integrating this pipeline into any risk management system, governments around the world can have a more structured and streamlined approach to fraud detection and increase the efficiency of fraud analysis and government revenue. Smart Fraud Detection The ML solution is built on the data from historical transactions of customs and provides predictions for newly arriving declarations, possibly even before the goods arrive at the border. It identifies not only potential fraud, but also non-risky items. The definition of fraud can vary from country to country, so the model can be customized to support various types of fraud as needed. It also provides an explanation for each prediction that includes the reasons behind its conclusion. To further help the officers make data-driven decisions, the solution identifies similar transactions from historical data and displays it on the dashboard. With all this AI-derived knowledge, the officers can confidently decide which items to inspect and how to manage fraud per declaration. Fake Invoice Detection The Fake Invoice Detection solution uses image processing techniques to identify fraud based on the authenticity of the document. The solution compares the incoming document's layout with past documents from the same vendor, provides a similarity percentage, and highlights the areas on the document that seem suspicious. Furthermore, the solution analyzes the metadata of the document, providing additional information on its creation and modification history. Smart Criteria Evaluation and Generation This product analyzes the efficiency of human-created risk criteria for customs declarations and uses AI to generate criteria that are, on average, 3 times more efficient. The solution provides officers with a list of efficient AI-suggested criteria and notifies them when an active criterion is not providing the best results. All this information is then displayed on an interactive dashboard, allowing the users to enable/disable certain criteria, see efficiency scores, the projected number of fraud hits, etc.

Results

Portmind’s risk management ML modules provide a holistic approach to fraud detection. They analyze the different components of the transaction, such as visual, textual, and tabular data, to get a comprehensive understanding of riskiness of the transaction. By integrating these modules within the risk management systems, governments decreased the number of non-fraud manual inspections by 30-40%, depending on the quality of data. They also received criteria that were 3 times more efficient than the human-generated ones. The product also can also provide risk scores in real-time for any entity or event of interest. All this data is presented to the user in structured dashboards, where they can monitor the AI insights and performance indicators across time. To help users understand the AI predictions and make data-driven decisions, the solution provides explanations and retrieves similar transactions from historical data. In order to adapt to changing conditions and new actors, the ML models behind the Risk Management Assistant constantly learn from new data to identify new kinds of fraud and anomalous behaviors.

Want to see how the solution works in practice?