Can Transformers Transform WCO Data Model Interoperability?

A Transformer‑Based, Human‑in‑the‑Loop Pipeline: A Design‑Science Ex‑Ante Evaluation

Abstract

“Interoperability” is never merely a slogan - but the very practical feature in all inter-organizational IT solutions and regardless of whether AI is involved.  Public sectors, such as customs administrations and their partners, must exchange the same facts with the same definition across declarations, manifests, guarantees, and post-clearance audits. The World Customs Organization (WCO) Data Model (DM) is a common dictionary for that work, published via the eHandbook and the DM App. However, day-to-day implementation still struggles with semantic heterogeneity, version drift, code-list changes, and the need for auditable decisions. This study specifies a single, governed design that addresses these realities together through a transformer-assisted mapping pipeline that retrieves candidate DM elements from authoritative text; re-ranks them with a cross-encoder for precision; enforces hard constraints on data types, cardinalities, and code lists; and routes uncertain cases for human review. The artifact is evaluated ex-ante using design-science method requirements, traceability standards for conformance to WCO Data Model Standards, alignment of risk-control registers with public-sector AI expectations, and scenario walkthroughs. A separate impact section compares life before and after adoption. This study also explains why transformers are a fit for language‑centric schema alignment, while being clear about their limitations along with the safeguards needed. Examples used in this study are from publicly available sources only; no proprietary data was utilized.