AI and customs – automation and machine learning

This article is a continuation of a previous research piece on customs and artificial intelligence, that introduced the Harmonized System, created to facilitate trade, harmonising it by using commodity descriptions and the corresponding coding system.

AI and customs – keeping it accurate

When looking into the customs trends we will see this year, it is obvious that customs clearance is moving towards a higher digitalisation, which is already benefitting businesses importing and exporting across the globe, but a lot of progress still needs to be made on artificial intelligence.

When looking into the varying degrees of automation existing today in the area of tariff classification there are several tools that can support. Currently, these can be largely grouped into three categories: simple systems based on keyword search, more sophisticated systems with algorithmic analysis of classification options (sometimes called expert systems) - which are in a way forerunners of machine learning (ML) - and, finally, various AI and ML systems. What are these? What can be identified as their strengths and weaknesses? How will they serve customs and HS users? To start, let us dive into the possibilities that these tools can bring to this industry.

Precursors of tariff classification automation

Among the different systems that exist today, expert communities identify:

1. Keyword search systems: In existence since the 1990s, they are defined by Merriam-Webster as “computer software used to search data (such as text or a database) for specified information”. The most used example today is Google. Search engine tools for tariff classification are helpful for searching for specific terms to retrieve HS codes and see where a given product could potentially fall. One of those engines is what is widely known as WCO Trade Tools. The engine allows users to search through the HS Nomenclature, the Explanatory Notes and the Classification Opinions for a specific term. The user introduces a particular term, e.g., “milk”, and all references with that term are shown, including “milking machines”, etc. Search engines sometimes provide users with hundreds of references connected to the searched word, which they then need to analyse to determine which one of those is correct. While it is facilitating the classification process, which could otherwise result in a lengthy search, the usefulness of such engines has certain limits. The main disadvantage of such tools is that in order to arrive at the correct result users still need to know how to find their way through dozens of options, and know how to use legal notes, how to apply General Interpretive Rules (GIRs) to classify something, etc. To add to the complexity, in the HS product are not necessarily named in the same way as people call them in their day-to-day life. An example is the term “computer”. The engine will not find a matching code for that term simply because in the HS computers are called “automatic data processing machines”. So, these nuances remain a limitation, ultimately making search engines great for speed but still a tool to be used by experts.
2. Expert systems: Invented in the 1970s, expert systems mimic the knowledge and expertise of a human expert in a particular field using rules and algorithms to provide complex problems with solutions. These are usually used for diagnosis, decision-making, prediction, and process automation. To use these systems in connection with customs clearance and commodity classification in particular, users don’t need to have any specific preliminary knowledge of classification rules. The system guides users through the algorithms and prompts them to answer specific questions about the product (asking for characteristics) so that their search can be accurate and produce one single commodity code at the end. An example? The word “wine”. The expert system will ask questions such as: “Is it flavoured? Is it sparkling?”, thereby narrowing down the choice of options. In some cases, this process may be slightly longer than the use of search engines, but here even users with little knowledge of the HS can successfully use the system. They just need to have detailed information on the product they want to classify.

So, given these two systems, how can a third option – artificial intelligence and machine learning algorithms – unite and boost their capabilities to fill the existing gaps?

AI customs clearance automation and machine learning

At the WCO Conference on the Future of the Harmonized System (HS) back in 2019, one of the main topics in focus was the complexity of the HS in regard to the challenges experienced by companies in using the right tools to consult and manage the data from the HS. To overcome this complexity, today tools are “made easier to use” with the help of AI and machine learning and they are being developed and trialled by a growing number of companies, including customs compliance. The AI systems that are in development to support this industry, learn from the data that they are fed with, processing it to “learn” and improve progressively. These wouldn’t follow a specific programme but would work based on a “statistical likelihood”, automatically assessing the chances of a product being classified in a specific subclass, or in several different subclasses, giving the likelihood of a certain classification option (70% this option, 10% this one, 30% another, etc.). This is how the algorithm works, and one of the advantages is that it provides different options, allowing the users to choose. At the moment, some basic knowledge of the HS classification is still required to make the choice. Although the system may put 97% on a specific classification, a custom and HS expert is able to correct it and make the system learn.

Blending different systems into one “hybrid” tool would probably be the most advantageous solution for logistics providers, service providers, importers, exporters, and traders. This is because it will incorporate all the advantages of search engines (to quickly find references across the HS tools and tariffs) and expert systems (to handle more complex cases requiring manual checks), layered with AI and ML functionalities on top (to boost these systems with a smart and automated extra power), finding the correct match with speed and accuracy.

Just like in all other industries, there is still a great amount of scepticism about machine learning and AI in the context of commodity classification. The key to overcome that is to make sure that the knowledge base used to drive the machine learning is properly curated and contains only correct data. “It is really critical that when building customs machine learning tools these focus on the integrity of the data that goes into their knowledge base” reiterated William Petty, Global Product Development Manager within Global Trade & Customs Consulting at Maersk. As the technology improves, the commodity classification expert community will be able to rely on it more and more to get the desired results.

AI customs clearance automation and logistic providers

How can logistics providers use AI, with the right balance of machine learning and human intervention, applying it to each individual case? For companies, adopting the right degree of automation for tariff classification can be quite a challenge. There can be so many different options, involving individual exporters/ importers, functioning in totally different ways, and dealing with different kinds of commodities. Going forward, ensuring the compliant and uniform classification can be an overwhelming struggle, so it is very important to understand that these AI tools are there to support and not to replace experts. People involved in the classification should be trained and updated on recent developments in this field. Automation needs to be approached in a very careful way, and logistics providers, particularly those offering customs clearance and classification services, will need to ensure that AI is leveraging good data (rather than mass data), and that companies are equipped with robust training programmes. Practically, logistics providers can support their customers with these processes, helping them classify goods and/or deploy further automation in a smart, selective, and efficient manner.

In summary, how can AI boost customs?

Artificial intelligence and machine learning certainly have the opportunity to revolutionise commodity classification and customs clearance, but their learning needs to be based on correct, highly accurate, approved data. Here below is a summary of the pros and cons.

Pros

• Acceleration of the classification process: finding the correct tariff codes more swiftly, which particularly benefits traders handling large volumes of heterogenous products to be shipped.
• Stronger tech support: useful for HS experts to make their work more efficient and for non-experts to help them gain knowledge.
• Continuous improvement: the more the AI program is used, the more powerful it becomes, boosted by the incremental learning of the software.

Cons

• Limited accuracy: the AI still needs to “learn” to provide accurate results, so 100% accuracy of classification cannot yet be ensured yet.
• Scepticism: The functioning of the AI is still a not fully understood. The algorithms are yet to be further adjusted so their use can still encounter some aversion among HS users. There is still a high likelihood of AI-generated errors, so users need to be aware of that in terms of customs compliance.

“Artificial intelligence tools, or any automation used for commodity classification, won’t go against the spirit of the HS - which is primarily about trade facilitation – but rather goes hand in hand with it. By introducing automation, we are supporting the main promise of the HS of streamlining the global commerce, adding another powerful layer enabling its efficient implementation and use” says Alexey Shcheglov, Product Lead Expert for Global Trade and Customs Consulting at Maersk. Businesses can use AI-boosted HS tools to their advantage, as part of their import and export strategies, which also helps to approach more strategically their supply chain’s opportunities. A sort of balanced dance between duties and growth.

Note: this article has been completely written by a human.