Automated Customs Code Attribution – DJN

Challenge

Jan De Nul, a global expert in maritime infrastructure, faces ongoing logistical complexity in transporting materials and parts across international borders. Each shipment requires assigning a Harmonized System (HS) code, which determines the applicable customs duties. Because code attribution varies by both product type and destination country, and descriptions differ between internal catalogs and customs databases, selecting the correct code is a time-consuming, high-stakes task. Errors can lead to overpayment, regulatory fines, or blocked shipments—delaying critical operations and impacting customer timelines.

Nature of collaboration

To address this challenge, Jan De Nul partnered with Sagacify to build a solution that could automatically assign HS codes with a high degree of accuracy. The collaboration focused on creating a dual-model system tailored to the complexity of customs classification, while factoring in business risk to determine when automation is appropriate.

What we built

Sagacify developed a two-stage AI system designed to process technical product data—text, images, or both—and recommend the most relevant HS code. The solution combines two models:

• A retriever model that filters the customs database to find all potentially relevant HS code sections and subsections based on the product’s description
• A reader model that evaluates these candidates and selects the best match based on context

The system accounts for the risk of misclassification by deciding whether to automate the decision or defer to human input. It also maps internal product codes and descriptions to the customs taxonomy, bridging the language gap between internal and regulatory documentation.

Impact