Amid the active engagement of Members in the BACUDA project through the development of their own version of analytic models, Azerbaijan Customs shared encouraging news for Members’ journey toward a data-driven organization.
Under the guidance of the Chairman of the Azerbaijan Customs and inspired by the BACUDA algorithms, A graduate from the BACUDA Scholarship program at Azerbaijan Customs has created an advanced AI Harmonized System (HS) tool. Utilizing the BACUDA open-source model as a base freely available on the CLiKC!, the tool introduces innovative features that augment its utility and efficiency.
The model has been trained on over 1.6 million Azerbaijan Customs data points since 2022, achieving an impressive accuracy of approximately 85% for 6-digit HS codes and 82% for 10-digit codes. Notably, it even exceeds 95% accuracy for 11 specific chapters where more data is available. Enhancing the model's accuracy is a critical focus of its ongoing development. This will be achieved by enriching it with large volumes of high-quality data, particularly in the chapters where current accuracy levels are suboptimal.
Key Features of AI HS Tool
Multi-Language Support: The tool incorporates Google's translation service to auto-detect and translate queries, allowing users to search commercial descriptions in multiple languages and still obtain identical results. This feature expands the tool's reach on a global scale, potentially simplifying the process for trade brokers worldwide to accurately identify classifications in accordance with Azerbaijan's regulations.
Chat-GPT Integration: The tool supplements its AI-generated recommendations with insights from Chat-GPT via an API. Since only non-sensitive commercial descriptions are shared, user privacy is upheld, in compliance with data protection regulations. This dual-recommendation approach allows users to cross-verify results. If one system makes an error, the other can serve as a check, enhancing reliability. For example, AI HS accurately classifies 'white chocolate' (1704903000), whereas ChatGPT recommends the HS Code used for chocolate (1806200000).
Duty Rate Calculator: By clicking a calculator icon next to the recommended HS code, users can input additional data to instantly compute duty rates. This functionality enables quick rate comparisons for the current HS Code and aids in ensuring compliance with customs regulations.
Risk Management Integration: The tool is integrated with a risk management engine to audit and validate customs classifications. It flags incorrect classifications associated with lower duty rates, facilitating data-driven decision-making and resource allocation for inspections. By flagging irregularities, the tool acts as an ongoing compliance monitor, which could deter fraudulent declarations and facilitate corrective actions. Applying this automated auditing function, the tool has already identified multiple misclassifications in their historical data.
These enhanced features collectively aim to create a more efficient, reliable, and user-friendly customs environment, thereby contributing significantly to trade facilitation and regulatory compliance.
We place significant importance on highlighting that the AI is designed to augment, not replace, the expertise of customs officers in classification tasks. While the tool's accuracy metrics indicate it is reliable, it is intended to serve as a supplementary opinion rather than a final arbiter in classification decisions. As global trade volumes surge, yielding hundreds of thousands of cases daily, this rapid and automated recommendation and auditing tool aims to bolster the efficiency of customs officers, enabling them to manage their workflows more effectively.
We are committed to cultivating a global network of Customs experts and will continue to support similar initiatives. Following suit, Cambodia, Maldives, and Peru Customs have also developed their AI HS tools and are on the verge of integrating these modules into their respective E-Customs systems.
Azerbaijan Customs' Approach: Navigating Discrepancies in HS Codes and Commercial Descriptions
Azerbaijan Customs recognizes the pivotal role of the AI-HS tool not just in detecting inconsistencies between HS codes and commercial descriptions, but also in its broader potential for facilitating accurate trade documentation.
It is imperative to note that when the AI-HS tool identifies an HS code that doesn’t align with its recommendation, it does not necessarily imply an error in the chosen HS code. Often, the discrepancies might be rooted in the commercial descriptions of goods, which may not have been sufficiently detailed or might have been miscommunicated.
By harnessing the capabilities of the AI-HS tool, Azerbaijan Customs is poised to achieve two-fold benefits: ensuring accurate HS code classification and refining the quality and accuracy of commercial descriptions. This proactive approach aims to create a more streamlined and precise customs documentation process, bolstering the confidence of both customs officials and traders in the system.
Future actions to take
The project team is committed to promoting the potential of human and technology collaboration and continue to provide technical assistance and practical workshops on the way Customs officials are able to improve their daily work performances by adopting machine learning and AI for Customs purposes.
To support Members in better understanding and embedding data analytics in Customs organizations, the BACUDA Project is launching advanced online courses on data quality and image recognition soon on the CLiKC! Platform.
Technical support
For further customized support, WCO invites Members to contact the WCO BACUDA project team (bacuda@wcoomd.org) to organize a joint test of the HS Recommendation AI model with BACUDA experts.
For more information, please contact capacity.building@wcoomd.org.