Structure Without Schema Is Not a Solution

The AI pitch in customs is a compelling one: ingest a scanned packing list, a PDF invoice, a free-text email — and produce structured output. The claim is technically accurate. A model can do this. The output is real.

But structured without a declared schema is just a different format of ambiguity.

Here is the precise failure mode. When every actor in the supply chain uses AI to impose its own structure on unstructured inputs, you get N locally coherent outputs that cannot be joined without another inference step. The chaos does not disappear — it migrates upstream into the integration layer. You have traded document chaos for schema chaos.

The commercial consequence is direct. An AI system that structures its own output may pass a compliance check at the border. It will not build the behavioural track record that a risk engine needs to accumulate across thousands of shipments into a trusted trader profile. That requires not just structure — but the same structure, consistently, across every actor in the chain.

Structure without schema interoperability is compliance theatre.

Global trade technology needs a declared, interoperable standard that every actor can conform to — so that the output one system produces can be joined, verified, and trusted by the next. Not a better AI. A standard schema.

The Customs Canon's Trade Line Assertion schema is a direct response to this problem. The TLA is a canonical junction object: a machine-readable declaration of the relationship between a commercial invoice line item, the physical piece or pieces containing the invoiced goods, and the digital product credential carrying that piece's regulatory attributes. It is the smallest indivisible unit of customs-grade trade data at which genuine AI reconciliation becomes possible — precisely because it declares its schema rather than inferring it.

That problem is solvable. It is not being solved fast enough.

First published on LinkedIn, May 2026. Greg Pilkington is a Global Customs Advisor and author of the Global Customs Canon, published under CC BY 4.0. and available at gctforum.org/gcc.