As global supply chains become increasingly complex and tariff landscapes shift unpredictably, the traditional approach to Harmonized System (HS) product classification is breaking down. IDC research reveals that companies using manual classification processes spend 60% more time on compliance activities. Here’s how we got here — and where the industry is headed.
Jump to ↓
Era 1: Physical inspection and printed tariff schedules (Pre-1988)
Era 2: The harmonized system changes everything (1988)
Era 3: Digitization and the electronic filing era (1990s)
Era 4: Rule-based expert systems arrive (Late 1990s–Early 2000s)
Era 5: Machine learning enters the picture (2010s)
Era 6: The AI era — Classification as a strategic function (2018–Present)
Why ONESOURCE Global Trade is built for AI-powered classification
The path forward
Era 1: Physical inspection and printed tariff schedules (Pre-1988)
For most of modern trade history, classification was a hands-on, highly localized endeavor. Customs officials and trade specialists classified goods by physically inspecting them and matching them against printed tariff schedules — thick, country-specific volumes that bore little resemblance to one another across borders. Each nation maintained its own nomenclature, making international trade a patchwork of incompatible codes and interpretations.
Classification depended almost entirely on individual expertise. Errors were common, costly, and slow to resolve. A misclassified shipment could mean delays at the border, unexpected duty assessments, or regulatory penalties — with limited recourse and no shared standard to appeal to. For trade professionals of that era, classification was less a system than an art form passed down through experience.
Era 2: The harmonized system changes everything (1988)
In 1988, the World Customs Organization introduced the Harmonized Commodity Description and Coding System (HS) — a single six-digit code framework adopted across more than 200 countries. It was the first major standardization breakthrough in trade history, giving the global trade community a common language for the first time.
But the process remained manual. Traders and customs brokers worked from paper-bound HS schedules, cross-referencing classification rulings and explanatory notes by hand. The shared framework reduced errors caused by nomenclature mismatches, but classification accuracy still depended on the knowledge and judgment of individual specialists.
Era 3: Digitization and the electronic filing era (1990s)
The 1990s brought the first meaningful efficiency gains. Customs agencies began digitizing tariff schedules, and early electronic filing systems — such as the U.S. Customs and Border Protection’s Automated Commercial System — began to emerge. Software tools allowed classifiers to search for codes electronically, and companies started building internal databases of pre-classified items.
The process was faster, but still fundamentally human-driven. The computer was a lookup tool, not a decision-making engine. Trade professionals were still responsible for interpreting ambiguous product descriptions, navigating multi-component goods, and applying the General Rules of Interpretation correctly.
Era 4: Rule-based expert systems arrive (Late 1990s–Early 2000s)
The first wave of true human-in-the-loop acceleration came through decision-tree and expert-system software. These tools encoded the General Rules of Interpretation into structured question flows — a classifier would work through a series of branching prompts and be guided toward an HS code recommendation. Early global trade management (GTM) platforms began to emerge during this period.
Accuracy improved, but the systems were brittle. They performed well for straightforward, well-described products and struggled with anything novel, technically complex, or ambiguously worded. Non-English product descriptions were especially problematic. These tools reduced the burden on individual classifiers, but they couldn’t replace human judgment in the long tail of difficult cases.
Era 5: Machine learning enters the picture (2010s)
As companies accumulated large historical datasets of classified shipments, classification began to be reframed as a text categorization problem: given a product description, it predicts the HS code. Early machine learning models — TF-IDF, Naive Bayes — gave way to more powerful techniques like gradient boosting and support vector machines.
For common, well-described goods in major trading languages, accuracy improved significantly. But the models still had meaningful blind spots: long-tail products, technical specifications in non-English languages, and multi-component assemblies continued to pose challenges. The models learned from patterns in historical data; they couldn’t reason about unfamiliar inputs the way a seasoned trade specialist could.
Era 6: The AI era — Classification as a strategic function (2018–Present)
The arrival of transformer-based architectures and large language models (LLMs) marked a genuine turning point. These models understand nuanced product descriptions, handle multilingual inputs natively, reason across multiple product attributes simultaneously, and can leverage retrieval-augmented approaches that pull in HS explanatory notes and prior customs rulings in real time.
Today’s AI classification tools can suggest a code with confidence scores, flag ambiguous cases for human review, and continuously improve from corrections over time. Several customs agencies are piloting AI-assisted systems for automated declarations. For trade professionals, this shift is not incremental — it’s transformational.
And the timing couldn’t be more critical. According to IDC’s 2025 MarketScape: Worldwide Global Trade Management Applications for Manufacturers and Exporters, the global trade environment has become “a dynamic, fast-changing environment, forcing manufacturers and exporters to develop the capabilities to more capably manage the intricacies of this element of their business.” IDC data shows that companies are overpaying duties by approximately 5%, and 20% of shipment delays can be attributed to inaccurate or incomplete customs preparations. Misclassification isn’t just a compliance problem — it’s a profitability problem.
The IDC report also highlights that AI and machine learning capabilities are being embedded into GTM software to help teams “more consistently, comprehensively, and accurately manage the dynamics of this environment” — with product classification cited as one of the leading use cases for delivering real AI value in global trade.
Why ONESOURCE Global Trade is built for AI-powered classification
This is exactly the problem that Thomson Reuters ONESOURCE Global Trade is designed to solve. Named a Leader in the 2025 IDC MarketScape for Worldwide Global Trade Management Applications for Manufacturers and Exporters, ONESOURCE Global Trade brings together AI-powered classification, comprehensive trade content, and a connected ecosystem — all on a modern SaaS platform.
The platform’s CoCounsel GenAI assistant streamlines classification workflows and delivers holistic insights across trade functions, helping global trade professionals move from reactive compliance to proactive risk management. A dedicated team of 200 global content specialists supports 220 countries and territories with more than 155 million content updates annually — typically within 24 hours — ensuring that AI-assisted classification decisions are grounded in current, authoritative data.
For global trade professionals managing an increasingly volatile tariff landscape, that combination matters. As IDC Research Director Travis Eide put it:
“Generating comprehensive, timely, and accurate intelligence across the end-to-end supply chain allows organizations to evaluate trade-offs in real time to improve compliance, identify opportunities to control and reduce costs, and strengthen supplier relationships.”
ONESOURCE Global Trade is built to operationalize that intelligence — transforming classification from a time-consuming manual burden into a strategic competitive advantage.
The path forward
The evolution of trade classification spans six decades and six distinct eras, each shaped by the technology available and the complexity of the trade environment. Today, AI-powered solutions are closing the gap between the speed of global commerce and the capacity of compliance teams to keep up.
For trade professionals still relying on manual processes or legacy rule-based systems, the cost of inaction is rising. The technology to do better — faster, more accurately, and at scale — exists now.
Ready to see what AI-powered trade classification can do for your organization?
Learn more about how Thomson Reuters ONESOURCE Global Classification AI and Trade Research AI help leading manufacturers and exporters turn compliance complexity into competitive advantage.
