Trustworthy AI for Customs & Trade
Insights from stakeholders from the (inland) shipping industry
Introduction
This project surveys how Artificial Intelligence (AI), data analytics, and digitalisation can strengthen customs supervision and trade compliance. Led by Erasmus UPT with support from AI Port Center, the work synthesises technology pilots (e.g., chat bots, X-ray image analysis), data-quality initiatives, and stakeholder dialogues (brokers, software providers, and authorities). The original scope—chat bots in Dutch Customs—expanded to a broader look at AI in customs due to changing circumstances, including parallel in-house developments at Dutch Customs
Insights from Stakeholders
Perceptions of “Smart Customs”: Practitioners see clear benefits in AI for productivity, consistent decision support, and multilingual guidance—provided answers remain legally exact and traceable to source law. Solutions like Picoco separate a curated “knowledge box” of regulations (database search for exactness) from an LLM that formats answers in any EU language, a pattern stakeholders rate highly for reliability. Lithuania’s “Matas” assistant has operated on this architecture since 31 Dec 2024.
Pain Points Addressed by AI: Frequent issues include repetitive Q&A to customer services, error-prone declarations, and resource-intensive screening of images/documents. AI can mitigate these via chat-based self-service, automated error detection, and image recognition that triages scans before human review.
Current Digital Maturity: Capability is uneven. Some customs agencies advance X-ray autodetection and restricted internal chat-bot pilots; brokers and software vendors explore AI features but often lack robust data pipelines and telemetry (timestamps, process metadata) to fully enable prevention-over-correction.
Barriers and Compliance
Integration Barriers:
Legal consistency & hallucinations: In a legalistic domain, non-deterministic answers are unacceptable; early tests showed hallucinations even with vetted sources, pushing pilots to start “inside the house” before opening to traders.
Org change & skills: AI shifts supervision concepts (e.g., first-line work by business, second/third-line by customs) and demands new team structures and controls.
Data gaps: Declaration data quality is variable; missing process signals (e.g., turnaround times) hamper detection complexity estimates and impact measurement.
Regulatory Alignment & Explainability:
EU reform foregrounds a Customs Data Hub and ML-enabled risk analytics, paired with human intervention for a 360° supply-chain view. Emphasis is on data quality, cross-validation, and rationalising data elements. Explainability must link every AI output to authoritative sources and auditable rules, consistent with EU-level guidance and the “Trust & Check” simplifications for fully transparent traders
Outlook
AI adoption in customs is accelerating—chat bots and image analytics are moving from pilots to practice, and EU reform will amplify data-driven supervision. Next steps:
Explainability & trust: Codify templates and decision rationales; expose rule checks alongside AI suggestions to satisfy auditors and traders.
Stakeholder integration: Engage brokers, software providers, and traders to co-define data-quality rules and service models that prevent errors upstream.
Digital Product Passports (DPP): As ESPR matures, DPPs could supply authoritative product data for multiple regimes (e.g., CSRD, EUDR) and eventually interface with customs systems—potentially the “real” customs revision leap when combined with the EU Data Hub.initiatives will ensure sustained competitive advantage and regulatory alignment.