evidoria

← Back to browse

Good practice Imported

CRMS — Nepal Customs' Machine-Learning Risk Targeting Now Covers 90% of Commercial Trade

Nepal · Kathmandu · See the Nepal profile · See the Kathmandu profile

Evidence: Descriptive / self-reported Top 24% 73/100 · Ask Evidence Copilot about this practice

Nepal Customs built an in-house machine-learning module inside ASYCUDA World, now running in the six largest customs offices, screening ~90% of commercial consignments and lifting revenue on targeted product lines by ~3.74%, per the World Customs Organization.

~3.74 %
Additional revenue generated on targeted product lines (current)
~90 %
Commercial consignments covered by CRMS offices (current)
30 indicators
Risk indicators examined by CRMS (current)
6 offices
Largest Customs offices running CRMS (current)
CRMS — Nepal Customs' Machine-Learning Risk Targeting Now Covers 90% of Commercial Trade

Details

Maturity
Scaling
Promoter
Department of Customs, Nepal (Ministry of Finance)
Period
2023–2026
Keywords
customs, trade facilitation, revenue administration, risk management

Context

Since 2016, Nepal's Department of Customs had relied on ASYCUDA World's built-in selectivity engine, which used static rules, manual risk profiling and officer discretion; internal case reviews found that some high-risk transactions were only caught after clearance instead of before it.

Objectives

To close that gap, a multidisciplinary in-house team of Customs, risk-management and IT specialists built the Customs Risk Management System (CRMS), a machine-learning module integrated into ASYCUDA World.

Activities

CRMS examines 30 risk indicators to flag misdeclaration, under-invoicing and over-invoicing, sends its recommendations to Customs officers for assessment, and is refined continuously as post-clearance audit and inspection results are fed back into its risk parameters.

Results

CRMS is now running in Nepal's six largest Customs offices, which together process roughly 90% of the country's commercial consignments. Reporting in the World Customs Organization's own magazine (March 2026), Nepal Customs said the system identified misclassification in a small share of targeted declarations and generated approximately 3.74% additional revenue from the product lines it targeted, while officers reported more consistent, better-supported decisions.

Conclusions

These figures are Nepal Customs' own reporting to the WCO rather than an independently audited evaluation, and the WCO article does not name the specific algorithm family behind the risk-scoring models. Nepal Customs has said future versions will add non-intrusive-inspection image analysis and deep learning to help detect concealed items.

Implementation
Indicative cost
Low (< €50k)
Time to results
Medium (1–3 years) — Built in-house and rolled out across Nepal's six largest Customs offices by 2026; future versions are planned to add non-intrusive-inspection image analysis and deep learning.
Staffing & skills
In-house multidisciplinary team of Customs, risk-management and IT specialists, Customs officers who assess each flagged recommendation, Post-clearance audit and inspection teams whose findings feed back into the risk model

Conditions for success

  • Integration with the existing ASYCUDA World platform rather than building a stand-alone system
  • A continuous feedback loop from post-clearance audit and inspection results back into the model's risk parameters
  • Human officer review and assessment of every flagged recommendation

Common failure modes

  • Figures are Nepal Customs' own reporting to the WCO rather than an independently audited evaluation.
  • The WCO article does not name the specific algorithm family behind the risk-scoring models.

Where it fits

Governance type
national customs / revenue authority
Scale
national (six largest offices, ~90% of commercial trade)
Income level
lower-middle-income

Commonly funded by

National / regional programmes

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful