International Trade is complex and often quite meticulous. The industry continues to grow more and more complex as technology advances and relationships among countries evolve. Continuous advances in technology force industry leaders to consider things they have never considered before. In 1990, world trade leaders probably were not debating where we should classify electronic waste. As the Harmonized Tariff Schedule (HTS) evolves, we long for a method of simplifying its use.
In 2016, a Thomson Reuters/KPMG survey found that 95% of respondents reported having difficulty classifying products with the HTS. Utilizing the HTS does take patience, critical thinking skills and extreme attention to detail. Categorizing products may seem like a simple process; however, when there are over 19,000 options and descriptions, it can be difficult to decide exactly where to place your “simple” jar of salsa. A few companies within the industry have decided to try and automate this tedious process with the use of artificial intelligence.
Artificial intelligence (AI) has become a top trend in seemingly every industry. From Tony Stark’s Jarvis to Amazon’s Alexa, AI went from a farfetched Sci-Fi movie plot to becoming a guest in all of our homes. In an industry with so many rules of interpretation, could AI be a viable option for classification in International Trade?
A successful type of artificial intelligence, machine learning (ML) learns to predict outcomes based on previous data sets. ML leaves room for slight flexibility, as it does not require strict programmed rules to function. The program truly learns over time and gains “experience”, just as humans do. For those bold enough to try AI in the classification process, this type of AI seems to be the “best” option to automate one of the more difficult processes within the Trade Industry.
A few barriers arise when we consider using AI for classification. HTS Classification requires the ability to justify (or at least strongly debate) your result. On the surface, it may seem like there are multiple correct classifications for a product. This is where your patience would be tested. Each section and chapter have notes that should be used to guide you in the right direction and hopefully prevent you from getting too far down the wrong path. Many chapter notes include specific exclusions that would take months or years to “teach” the AI. Once the AI recognizes these notes and exclusions, it would need to be able to explain how it arrived at the classification, citing descriptions, notes and exclusions.
Another potential concern for using AI for classification is that while the HTS is only updated every 5 years, these changes occur overnight. While humans would begin utilizing the new PDF file the next day, we would have to reteach the machine with new additions, as well as the changes to previous codes. While AI may be intelligent, it probably would not process a major tariff change overnight.
Possibly the most difficult barrier to overcome is the inability to classify based on certain descriptions. When manufacturers are developing products, providing descriptions that perfectly fit the HTS is not on their top concerns. When classifying, we recognize synonyms, typos or when a description does not provide enough information. As humans, we are able to identify what other pieces of information we may need based on the possible codes identified and the categories the part may fit into. By automating this process, we miss an opportunity to ask the right questions on the first try. For anyone who has spent time trying to navigate an automated phone tree, we would love to avoid the abundance of wasted time that occurs when we merely need to ask an agent a simple question. Fighting with the automated message system through an AI while your shipment is waiting to be imported sounds like a disaster.
As we have seen technological advances multiply beyond belief throughout the last 30 years, we know that artificial intelligence will continue to adapt and grow. While current machines may not have the capabilities to argue for or against certain rules of classification, this could certainly be where the future of the trade industry is heading. For now, we can keep our handy 99 chapters of descriptions, section notes, chapter notes and subchapter notes at the tip of our fingers online. We can look forward to the day when Jarvis may be arguing with us over where to classify Iron Man’s armor.
Have a question? Contact Jordan Perri, Allyn International.
Subscribe to TaxConnections Blog
Enter your email address to subscribe to this blog and receive notifications of new posts by email.