Global trade is run by a wide range of regulations and laws. This gives no space for approximations and human errors. That is why using AI in global trade is the most reliable approach, and Eurora pioneered AI for cross-border compliance. But do you know how Artificial Intelligence works in this field?
What better way to learn about AI than directly asking an expert? We spoke with Kristi Helekivi, Head of Data Science at Eurora. She explained how AI works, why it is used in trade compliance, and how Eurora uses AI to provide intelligent Cross-Border Compliance solutions.
Kristi started in Eurora as a Data Scientist, and she is now the Head of Data Science team. She always loved numbers, analytics, and science. She studied Mathematical statistics at the University of Tartu and took IT subjects as a secondary profession. Kristi previously worked for almost six years in the field of clinical research, specializing in statistical data analysis, statistical programming, and good clinical practice.
Here is what she shared with us.
What is AI and is it different from machine learning?
Artificial intelligence (AI) is a branch of computer science that builds smart machines. They can perform tasks that typically require human intelligence. In most cases, AI can describe all kinds of rules and is able to adapt to any client. This would be almost impossible to achieve with a rule-based system. Precisely because we cannot collect all the rules or descriptions, and a rule-based system doesn’t learn. If a description has a typo or uses a synonym, a rule-based system needs maintenance. New rules should be generated from scratch. AI on its side has great adaptability.
Machine learning is a branch of AI, and it aims at imitating intelligent human behavior. It focuses on using data and algorithms to imitate the way humans learn to gradually improve its accuracy. The machine provides patterns at one specific moment: that is why it needs to be continuously trained. To be beneficial over time, the machine needs to be able to adapt the rules it learned.
Why using AI in the trade compliance field?
Using AI in trade compliance is more secure. Machinery evaluates objectively while people tend to be stuck in assumptions. It makes perfect sense if you think about how to allocate an HS code to an item.
Humans make human errors, and each person knows only a small part of the huge nomenclature world.
Kristi gives an example to illustrate how hard it is for humans to work on nomenclatures and HS codes. If a human only worked on the “plastics” chapter and a description “bowl” comes in, he will assign the “plastic bowl” code. However, the machine knows that bowls can be also made from wood, glass, stainless steel, or ivory.
As Kristi explained, machine learning can also improve and adapt, contrary to a rule-based system. Adapt to any new markets, new clients, or changes that often occur in the trade compliance world. For example, there are frequent HS code nomenclature changes. A human can forget about updates and use outdated versions. With AI, you only need to introduce the changes once.
Why does Eurora use AI?
Eurora uses AI for HS-code allocation. We can say that AI is at the heart of Eurora solutions because all the services work with the HS code allocated by the AI. If we take a step back, we can also say that HS code allocation is the heart of global trade compliance. Without an HS code, it is impossible for authorities to know what goods transit. Therefore, impossible to apply taxes and duties or to make sure those goods are compliant. Authorities also collect HS code data for statistical purposes so they can establish what type of goods are transported the most.
In the trade compliance field, people have been assigning codes manually for a very long time. However, one person cannot master the whole nomenclature. For example, knowing simultaneously clothes, live animals, electronics, and cosmetics sections. It also requires countless time and resources, which is not sustainable for businesses.
A machine has a wider horizon than one person. With AI, all the knowledge coming from different people is in one machine together.
Being aware of these challenges, Eurora developed AI for HS code allocation. AI prevents human errors, it’s a fact. Besides not being able to know every HS codes, humans are subjective. With AI, however, it is possible to provide the most accurate data but also handle large volumes. “AI has no vacation; it always does its job.”
How did Eurora develop AI?
“When we started to develop AI, we tried to capture human decision factors. We captured their life experience and the information coming from rule books.”
First and foremost, the machine needs data. Eurora’s AI studies all the product descriptions and HS code pairs using supervised learning. “We know which HS codes we would like the machine to be able to predict.” Based on this information, it can start searching for the patterns and the rules. The machine always tries to find and use the best rule from its existing world. Word embedding is also an important input. “We need to teach the language to the machine”. The machine learns a language based on the contextual similarity of words. It tries to understand the similarities and differences between words based on texts. For example, separating “cat”, “kitty”, “cats” from “elephant”.
In other words, Eurora’s AI input data is product descriptions, and output data is harmonized system tariff codes. And to make it work, there is no secret: it’s all about research. More than 100 Experts, scientists, analysts, and researchers have been involved in AI development. You need to “study existing solutions and solve tasks by researching scientific articles. But also investigate similar problem-solving in the field of machine learning.”
How does Eurora’s AI help its clients?
When a client uses the HS code allocation service, the AI provides the answer to the query in 0.39 ms. It is able to provide instant answers to queries to new customers as well. It is also able to handle large volumes. The AI is very adaptable and provides fast and accurate data. It can generalize if there is not enough customer-specific data yet, or handle customer specifics. This flexibility wouldn’t be achievable with a rule-based system.
It is also interesting to know that beyond allocating the HS code itself, AI has other beneficial features. AI makes sure the quality of the service offered is always high through quality controls. It can detect anomalies or problems on clients’ side in real-time and react right away. The AI can also save clients time thanks to binary models for process optimization. This means that the AI is able to know if the data provided by the client is not enough to provide an HS code. If this comes to happen, it will tell it to the customer right away.
How does Eurora plan to use AI in the future?
Kristi’s team uses regression models to meet the needs of clients. This allows them to know when larger volumes are coming in and adapt. They also use business analytics to understand customers’ issues. Improving the AI requires close communication with other teams too, like the customer support team. They make sure that customers give good descriptions to the machine, so it can become smarter. Currently, Kristi’s team is focusing on expanding the AI capability on HS code allocation. They work on different destination countries, as well as export control codes.
Our AI is learning every day and in continuous development.
As Kristi mentioned, improving Eurora’s AI is a daily challenge. It requires not only computing resources but also new competencies, in a large amount. And yet, all this background work, research, and analysis enabled Eurora to develop the world’s most advanced machine learning technology.
Eurora’s AI is not only helping HS code allocation processes but also Duty & Tax Calculation, IOSS, Restrictions Screening, and Customs Clearance services. Are you willing to automate your compliance processes with secure AI-based solutions? Get in touch with us!