Challenges of Automatic Classification

Throughout my 30-year career with U.S. Customs and Border Protection (CBP), from when I first began as an Import Specialist in Cleveland, Ohio, until retiring as the Executive Assistant Commissioner for the Office of Field Operations in Washington, D.C., CBP’s trade mission has remained incredibly complex and ever evolving.  Yet at its foundational level, ensuring trade compliance remains directly tied to the accuracy of the Harmonized Tariff Schedule (HTS) classification.  Not only duty rates, but more significantly, admissibility requirements attach at the HTS ten-digit classification.  Ascertaining the accurate HTS classification has always required formalized training, applying a methodical thought process referencing the General Rules of Interpretation (GRIs), Explanatory Notes (ENs), Chapter Notes, Customs Bulletins, previous Customs rulings, and other resources; experience is only developed over time.  Of late however, with the development of Artificial Intelligence (AI), tools are being developed which may enhance the accuracy of the HTS classifications for those performing these complex tasks.  

The application of AI as applied to automatic HTS classification faces many operational and practical challenges, from attempting to determine the proper HTS classification when the product descriptions are vague, incomplete, or misleading, to the lack of quality training data from which to develop and mature the necessary AI algorithms.  HTS GRIs, ENs, Chapter Notes and Rulings, which often play a critical role in determining proper classification, are difficult to incorporate into these automatic classification tools.  Additionally, trying to determine the proper HTS classification of new, unique and complex products can be a challenge for even the most experienced licensed Customs broker.

While acknowledging the challenges facing automatic classification technology, there must also be an awareness of the risks and pitfalls, from an eventual overreliance on technology, to a general distrust of opaque, black box solutions.  A changing external environment causing frequent revisions to the HTS must also be addressed as the consequences of inaccurate HTS classifications can be costly financially, impact a company’s profit margins, lead to regulatory actions, and may cause a delay (or refusal) to imported merchandise. 

Despite the challenges and pitfalls, the future of AI driven automatic HTS classification technology is promising, the use of which may be one component in helping an importer demonstrate that they have used “reasonable care” as they fulfilled their duties under Section 484 of the Tariff Act, as amended (19 U.S.C. 1484).  An importer’s failure to exercise reasonable care could delay the release of imported cargo and even lead to penalties or a referral for criminal enforcement.  Cognizant that the facts surrounding every importation differ, neither CBP nor the importing community has developed a clear reasonable care checklist which would cover every import transaction.  Instead, CBP has provided general guidelines to help importers in the exercise of reasonable care.  These guidelines include having access to the CBP regulations, HTS rules, Customs Bulletins, Customs rulings, and similar resources; having a responsible and knowledgeable individual within your organization overseeing the preparation of entry documentation; using experts to assist in your complying with customs regulations; having a customs compliance program; and ensuring that CBP has awareness when your identical transactions are handled differently at different ports of entry.  The use of AI driven automatic classification technology designed by experts and based on the HTS rules, Customs Bulletins, Customs Rulings and other resources may prove to be a powerful enhancement to an importer’s existing import documentation processes and may add further evidence that an importer has exercised reasonable care.

With that understanding, one should not assume that any AI driven automatic classification tool by itself can take the place of an appropriately trained, experienced and supervised individual.  The ideal end-state is for a complementary process which leverages the capabilities of AI with human interaction – humans to make the judgements, technology to process and synthesize information at lighting speed.  This approach is being applied today by CBP in their cargo security mission.  CBP has rapidly been moving to apply AI driven image analysis technology in their non-intrusive cargo inspections.  AI driven automated threat detection technology is being applied to all x-ray images while well trained and experienced CBP Officers apply their judgment to the review of the anomalous images.  This approach may be an effective roadmap for the importing community to follow as well.  By incorporating AI driven automatic classification technology into your entry document preparation processes, supported directly by human interaction which applies judgment and is performed under proper supervision, this approach may provide a considerable enhancement to your current processes, allowing for more accurate and timely HTS classification decisions, decisions which align with the reasonable care expectations of CBP and avoid classification errors leading to increased costs, fines, and delays to your importations.