ML scores declarations for inspection lanes, fraud and smuggling risk
- Who we serve
- Governments and Public Agencies
- Category
- PFM, Tax & Customs
- Value
- Risk management
- Human role
- Human approves every decision
- Maturity
- B-READY topics
- International Trade
The problem
Without customs risk targeting & selectivity, customs authorities rely on sampling and after-the-fact review, while the signals in tax returns, invoices, customs declarations, third-party and financial data sit unread, so problems surface only after the loss, incident or complaint.
How it works
Customs risk targeting & selectivity - ML scores declarations for inspection lanes, fraud and smuggling risk. It draws on tax returns, invoices, customs declarations, third-party and financial data and is typically owned by customs authorities. Where the human comes in: customs authorities review and approve before any action affects a person, a payment or a case; the AI flags, ranks and drafts. It is scaled at leading organisations and piloted elsewhere; examples include Jamaica, Barbados, Rwanda, Zimbabwe, Timor-Leste (ASYCUDA); Brazil SISAM.
Where the human comes in
Customs authorities review and approve before any action affects a person, a payment or a case; the AI flags, ranks and drafts
What it needs
tax returns, invoices, customs declarations, third-party and financial data
What to plan for
Human-on-the-loop for selectivity; false positives
In the field
Jamaica, Barbados, Rwanda, Zimbabwe, Timor-Leste (ASYCUDA); Brazil SISAM
UNCTAD ASYCUDA new generation (2025); WCO AI guidance (2023); IMF TNM 2024/006
Public deployments by other organizations, listed for context. Not Sphere projects.