AI in Tariff Classification: “People fear what they don't understand and hate what they can't conquer"
October 24, 2023
If you are reading this article there should be three reasons why you are doing so:
You are an AI sceptic, who doesn't believe that modern technology is capable of solving tariff classification or making reliable predictions.
You are open-minded and curious about the current trend of AI technology in tariff classification.
You are an international trader/company/CEO/team leader/service provider, etc. who is seeking a solution to a practical problem specifically customs classification.
Current situation and challenges in customs classification
Reflecting on many sceptical posts I aim to summarise the current state and actual challenges faced by the "world of customs classification."
Companies and service providers, among others, are facing struggles with customs classification (which is not limited to classification alone but extends to various customs-related matters) due to the following reasons:
Lack of expertise
Administrative overhead and bureaucratic burdens
Outdated/old-fashioned tariff language
Controversial interpretation of HS and national nomenclatures
Absence of various national language versions for customs classification decisions (such as HSEN, EU COM Minutes, etc.)
Managing large volumes of datasets
WCO and EU COM tools
To address the challenges in customs classification, the WCO and the EU COM regularly and irregularly publish classification information such as regulations, decisions, Binding Tariff Information (BTI), minutes, and more. They also provide open-source databases for reference. However, in a typical scenario where compliance is crucial, one must consult at least five or more different systems or databases to obtain the correct tariff number. Additionally, significant time must be allocated to interpreting these resources as the language used in these decisions, and even in the HS Nomenclature itself, is often legalistic and unfamiliar.
Indeed, classifying just one product can take a significant amount of time, easily exceeding hours. In the real world, companies and service providers involved in international trade handle hundreds or even thousands of SKUs on a daily basis, all of which require accurate classification.
If one is unfortunate and, for instance, customs classification decisions are either untranslated or unavailable in their country, it can create further obstacles in the process. In such situations, it is crucial to explore the existing solutions available. Here are some options to consider (whether helpful or not, you can decide):
Submit a BTI/Ruling application and wait for months until you receive a response from the relevant authority, providing additional information regarding the product you requested the BTI/Ruling for.
Submit a non-binding tariff information, which can undergo a legal process of review and potential denial by the authority.
Consult tax and legal advisors, incurring hourly rates of 200 EUR or potentially more, until you receive their consultation reply, which will provide additional information about the product you inquired about.
None of the options mentioned above, simply choose the tariff number associated with the worst-case scenario, which corresponds to the tariff number with the highest duty rate.
Digital approaches in customs classification
To obtain a clear understanding of typical digital approaches in customs classification, here is a concise summary:
Decision tree models
This represents a rule-based approach, employing a step-by-step classification diagram or model. However, due to the manual setup of rules, it is a time-consuming process and can only be applied to a limited number of products. Additionally, maintaining and updating the rules can be challenging. Nonetheless, this model can serve as a foundation for machine learning techniques. While it offers a high level of precision, its usability is restricted to specific products or product groups.
Machine Learning and Fuzzy matching
This approach can be categorised as a subset of AI, as it utilises training algorithms to learn patterns and make predictions. Fuzzy matching is a technique employed to compare texts or other data, such as article numbers, in order to assess their similarity. From a customs classification standpoint, this approach represents a semi-automated solution, where tariff number predictions are derived from historical hand-coded rules. However, it shares the same limitation as decision tree logic in terms of being reliant on predefined rules.
The latest technology in customs classification is deep learning, which falls under the umbrella of machine learning and utilises neural networks to analyse and learn from large datasets. Unlike traditional machine learning, deep learning operates with multiple layers (typically 3-4 layers) and has the ability to continuously analyse and monitor vast amounts of data. What makes deep learning particularly exciting is its capacity to automatically learn from datasets and make predictions without predefined limitations. Additionally, deep learning can be combined and integrated with machine learning and fuzzy matching techniques, allowing for synergistic effects and enhanced classification capabilities.
Our company, traide AI, is also utilising this technology. This means that the predicted tariff numbers are accompanied by justifications, specifically related to classification background.
Wrapping up: AI and human cooperation
Finally, I would like to emphasise a common misconception regarding the role of AI in customs classification. AI solutions are intended to assist and enhance the daily work of human beings. It involves a collaborative partnership between AI and human intelligence, aimed at reducing errors (although it may be hard to believe, experts and humans can also make mistakes) and ensuring compliance with legal requirements. The objective is not to replace human beings or experts, but rather to streamline and facilitate a process that is essential due to the challenges previously described.
There is a wealth of information and legal provisions to consider when it comes to classification. If the classification is incorrect, everything is incorrect: customs duties, export declarations, preferential treatment, export control provisions, excise taxes, sales taxes, and so on.
Why Choose traide AI?
At traide, we are dedicated to the customs tariff number. Our goal with traide is to offer an intelligent software package that achieves and maintains the completeness, currency, and integrity of an entire company's product master data (sometimes millions of products) according to customs tariff requirements over time. In addition, we address the issues that arise before the actual classification process and significantly influence it. This includes, in particular, the quality of the product description, which often provides inadequate information or presents this information in an unstructured and distributed manner (e.g., in HTML format on a webshop or in product data sheets).
At its core, traide software functions as an intelligent digital customs tariff expert. It helps to quickly find the correct customs tariff number, acting as a "Tariff Assistant." It also automatically checks already classified product inventories for formal and substantive accuracy and provides feedback to the user in case of errors. The result is significantly increased throughput and a lower error rate for classified products. This frees up staff resources, creates room for scaling, and enhances security.
Do you have any questions about this? Feel free to write to us! We are happy to assist you.