MyCustomsInfo®'s AI pipeline takes customs data from raw ingestion through a series of automated checks, then hands high value issues to human experts and finally outputs evidence ready audit packs and dashboards.
1. Ingestion and normalisation: The pipeline ingests customs declaration data (e.g. CDS/CHIEF exports, broker extracts, ERP files, and customs authority data feeds) plus supporting documents like invoices, packing lists and transport documents. AI components similar to FastNet transform unstructured and semi structured sources into clean, structured declaration records, so every data element (line item, HS code, origin, value, CPC, preference, authorisations, duty/VAT amounts) is normalised into a consistent schema.
2. Element by element AI audit: For each declaration, the portal runs AI and rules based checks across all elements, not just samples: HS codes, origins, values, preference/relief use, procedure codes, authorisations, and calculated duties/VAT. Models look for anomalies and patterns such as misclassifications, inconsistent use of preference codes versus origin and documentation, misaligned CPC vs. relief, suspect customs values, and duplicated or missing entries, assigning a risk/value score to each finding.
3. Human in the loop expert review: All AI flagged items are passed into a workflow for licensed customs specialists, who confirm or reject each issue and determine the correct treatment based on the applicable tariff, legislation and rulings. Experts enrich each confirmed issue with explanations, citations and recommended corrective actions (e.g. new HS code, corrected origin or value, revised CPC), turning AI suggestions into fully defensible audit records.
4. Outputs: actions, dashboards and learning: Validated findings are packaged into outputs such as duty recovery schedules, amendment packs, and remediation tasks, and pushed into dashboards that show overpayments, underpayments and compliance risk by entity, broker, country and time period. The system continuously ingests new declarations, re applies AI checks, and "learns" from expert decisions (for example on recurring product classifications), so future submissions require less manual intervention while every element remains auditable.