AI Use Cases in Customs Compliance: Are We There Yet?


Artificial Intelligence (AI) is permeating many areas of commerce, society, and governance. That includes the field of international trade — arguably one of the few fields where these aspects must seamlessly work together.

AI and cross-border trade

Experts predict that cross-border trade will reach a volume of several trillion USD by 2031. Today, many countries are dependent on imports and exports, and the free flow of goods and services is a major boost to many economies. However, these huge volumes of trade present a significant challenge to transport and customs authorities around the world. There are attempts to modernize the customs compliance systems to facilitate the process and reduce the administrative burden.

The HMRC’s upcoming upgrade of the UK’s 30-year-old customs compliance system is a big example of that. It’s no surprise that AI, being a major driving force behind overcoming many business challenges, has also made its way into international trade. I’d like to focus on three major ones and explore whether we’re on the way toward AI fully taking over the compliance process.

Customs classification

The process of classifying goods for the purposes of customs compliance in accordance with the Harmonized Tariff System is very complex. For example, if you’re selling men’s cotton T-shirts, you have a choice of several HS codes:

Image: Kate Sukhanova

Different measures, taxes, and duties will apply depending on your choice of HS code and the trade relations between your country and your country of import. You can easily lose money on customs fees — as far back as six years in some countries if you’re audited. A small number of companies proposed solutions, with the majority being AI-based. In this scenario, AI learns the customs declarations previously submitted. Here’s the problem with this: These data sets are inherently flawed.

First off, customs transactions that have already taken place act as the basis for the data sets, but there’s no guarantee these have been done correctly. There’s also no guarantee that an audit in three, or even five, years won’t completely overturn these declarations on the basis of a wrong HS code. The authorities can take a long time to overturn such declarations. However, AI systems that use them as a basis for learning have no way of classifying the goods without risking the overturns easily corrupting everything they “know” about specific commodities. According to Customs Compliance Direct, Mike Bing, “Relying on completed Customs transactions as a basis for validated tariff data is a dangerous quagmire, an area of quicksand which gives you the indication of standing on firm ground while being capable of swallowing you whole without leaving a trace.”

Secondly, the Harmonized Tariff System gets updated every five years. The update comes into force in the space of a day. There’s a high risk of the AI incorrectly classifying the goods under the new tariff system if it only knows the previously valid data sets. If the AI is updated with the new data sets, this can be corrected. Here’s an example of an AI tool’s customs classification provided by the World Customs Organization. I used the commodity “Axminster Carpets” to look up the code for goods made by one of the oldest carpet makers in the UK. These carpets have their own HS code, 5702311000, as provided by the HMRC:

The AI tool in question was unable to classify the Axminster carpets according to the existing commodity titled “Axminster Carpets”. It did provide a suggestion for the “Other carpets” commodity which includes Axminster carpets. However, I would’ve expected to see that option as an additional one to the Axminster Carpets commodity.

It’s not that AI can’t be used to optimize customs classification work. However, it’s not a 100% reliable tool, nor is it future-proof. The best way to combat the issue of customs classification is to employ AI together with customs experts. This should maximize software that integrates with governmental portals like the HMRC rather than software that’s fully reliant on AI.

Supply chain optimization

In the last few years, the world faced many logistical and supply chain challenges, and the sector has been dealing with huge pressure to manage the increased demand. With some challenges like Brexit and Russia’s invasion of Ukraine not showing many signs of stopping, the sector needs support offered by AI and technology more than ever.

OECD’s paper, published last year, stipulates that AI systems can support the industry by:

  • Helping with the management of inventories and warehouses through sensors and “smart warehouses”
  • Coordinating shipments across countries
  • Forecasting trends
  • Improving the accuracy of time-sensitive delivery

For the best decisions and output, smart warehouses should combine AI with tracking technologies like RFID, robotics, and IoT sensors.

Bonded warehouses

In terms of customs compliance, there’s no reason why AI can’t also support bonded warehouses. Before customs authorities process goods that go into free circulation, they are first stored in bonded or customs warehouses. I’d like to briefly mention the usage of data insights about the performance of such in different countries generated with AI and other technologies.

The UK is just one example of customs authorities operating in a silo rather than together with other countries. In fact, countries that don’t have trade deals in place with one another often operate that way without sharing information about the flow of goods between bonded warehouses and the efficiency of customs operations. Thus, businesses miss out on crucial data about customs operations in new markets they might want to break into. They also risk making the wrong decision about working on a bonded warehouse basis with some customers.

Having AI systems in place that can gather and process this data from different countries. This can also deliver actionable insights to both governments and businesses, allowing for better decision-making in the private sector about their international trade activities. It can also foster intergovernmental collaboration in terms of trade deals. Creating electronic customs systems that can work together can also allow mutual recognition of customs documentation generated by them.

Trade policy

AI has had a pivotal role in mitigating the outcome of unprecedented events. I’m not necessarily talking about international trade specifically. One of my favorite examples is actually in the MedTech field. BlueDot experts have conducted work on detecting COVID-19 spreads at the start of the pandemic with the help of AI. Their “warning system” can easily serve as an inspiration across other sectors – and international trade is one of them.

The OECD paper referenced earlier supports the assumption that AI can become a “useful tool to analyze real-time data and provide more timely and granular information to economic agents and policymakers, as well as help understand the impact of uncertainty or specific shocks.” With advanced AI tools in place, trade allies can work together to find mutually beneficial solutions. They can be proactive rather than reactive when such unforeseen events occur. In addition, AI can help governments make better choices regarding international trade legislation and support packages. For instance, it can provide real-time and overtime port data and supply chain bottlenecks. This helps export and import regulators determine which additional resources are needed.


In addition to the ones outlined, there are plenty of other applications for AI in international trade, such as:

  • ML-powered language translation
  • AI cargo scanning
  • Market intelligence and trade opportunities
  • Fraud Detection

As you can see, there’s definitely room for AI in international trade. However, we’re still a long way from AI fully taking over cross-border commerce and replacing logistics and customs experts. After all, not all intelligence is artificial, and that’s particularly true for international trade. To finish off, I’ll leave you with this quote from the recent Ecosystem of Trust Evaluation Report by the HMRC: “The specialist nature of these data items means there is little scope for full automation, although we noted a number of ‘helper’ facilities like look-up lists, detailed guidance, or Artificial Intelligence (AI) based initial calculations, for human review, aimed at reducing the effort required in this area.”

Photo credit: The feature image is symbolic and has been taken by Anek Suwannaphoom. The infographic and video has been created by the author for TechAcute.

Was this post helpful?

Kate Sukhanova
Kate Sukhanova
I’m a writer with a keen interest in digital technology and traveling. If I get to write about those two things at the same time, I’m the happiest person in the room. When I’m not scrolling through newsfeeds, traveling, or writing about it, I enjoy reading mystery novels, hanging out with my cat, and running my charity shop.
- Advertisment -
- Advertisment -
- Advertisment -
- Advertisment -
- Advertisment -
- Advertisment -