Tunisia's customs authority, known as the Diwan, is exploring the use of artificial intelligence (AI) and machine learning to analyze large amounts of data and target high-risk files. This initiative aims to enhance monitoring and decision-making, allowing for more efficient allocation of resources. The move was announced during a national meeting in the capital, where various projects and experiments were presented by officials from the Ministry of Finance and the General Administration of Customs.
The proposed AI system will enable the Diwan to identify high-risk files and focus monitoring efforts on priority cases. According to Essam Al-Fasatwi, director of risk management at the General Administration of Customs, a simulation using 2025 data showed that the AI model could detect an additional 96.9 million dinars, with a detection rate of 49.23%. The model was based on 11 million observations and 27 initial variables, which were later enriched with composite variables.
The AI model used a "decision tree" approach, allowing for clear explanations of the reasons behind classifying a particular file as high-risk. The Diwan's experience relies on integrating monitoring results into retraining and updating the model periodically. This approach enables a shift from static selectivity based on predefined rules to dynamic selectivity that benefits from data and previous monitoring results.
In the tax sector, the Ministry of Finance's information center is working on a project to use big data and machine learning to improve tax monitoring targeting. The project aims to identify high-risk files and allocate resources more effectively. The initiative also seeks to enhance the administration's ability to monitor the informal economy and new forms of digital activity.
The project, currently in the bidding stage, is expected to take a year to complete. The technical infrastructure will be based on local architecture and open-source tools, allowing for the use of experience gained in other public finance areas. According to Salih Al-Madab, director general of the Ministry of Finance's information center, this approach will enable better decision-making and more effective fraud detection.
The use of digital technologies has already led to significant reductions in physical and repetitive tasks, freeing up time for monitoring and analysis. With over 157,000 remote declarants and 87,000 landlord declarations submitted on time, the automated processing of landlord declarations has saved around 8,000 working days. However, the increasing volume of data from electronic invoicing and registered funds requires the development of new administrative capacities for processing and analysis.
The integration of AI can improve service quality, verify data consistency, and link transactions between various stakeholders. This will enable inspectors to focus on auditing, investigations, and analysis. However, this transformation requires the development of new skills among customs officials, particularly in statistics, data analysis, and understanding model outputs. The final decision on selecting files and intervening remains the responsibility of customs officials.
Key points
- Tunisia's customs authority plans to use AI to target high-risk files and enhance monitoring and decision-making.
- The proposed AI system has shown promising results, with a simulation detecting an additional 96.9 million dinars in high-risk files.
- The initiative aims to improve the efficiency of customs operations and enhance the country's ability to monitor the informal economy and new forms of digital activity.
