evidoria

← Back to browse

Good practice Imported

State Customs Committee of Uzbekistan Uses IBM watsonx.ai and Llama 3.3 to Automate 80,000 Customs Declarations

Uzbekistan · Tashkent · See the Uzbekistan profile · See the Tashkent profile

Evidence: Descriptive / self-reported Top 46% 60/100 · Ask Evidence Copilot about this practice

Uzbekistan's Customs Committee deployed IBM watsonx.ai with a Llama 3.3 model to auto-classify goods and process declarations, cutting per-declaration processing from 2-3 hours to seconds and fully automating over 80,000 declarations worth $5 billion in 2024.

80,000+
Declarations fully automated (2024)
$5 billion
Value of automated declarations (2024)
2-3 hours -> seconds
Processing time per declaration
$246,000
Additional revenue identified
State Customs Committee of Uzbekistan Uses IBM watsonx.ai and Llama 3.3 to Automate 80,000 Customs Declarations

Details

Maturity
Scaling
Promoter
State Customs Committee of the Republic of Uzbekistan (with IBM Expert Labs and IBM Client Engineering)
Period
2024-2025
Keywords
customs administration, trade facilitation, generative AI, public-sector productivity

Context

Facing a rising volume of customs declarations with complex free-text goods descriptions, Uzbekistan's State Customs Committee worked with IBM Expert Labs and IBM Client Engineering to build an AI system for automated goods classification and declaration processing.

Activities

The system uses IBM watsonx.ai and watsonx.data with a Llama 3.3 model extracting structured data from free-text fields and classifying products against Uzbekistan's customs codes; the Committee's IT footprint expanded from 9 information systems in 2021 to 37 in 2025, interconnected with 33 government agencies, and introduced 'AiRo', a passenger-facing customs-guidance robot at Tashkent airport.

Results

Per-declaration processing time fell from 2-3 hours to seconds; in 2024 more than 80,000 declarations worth $5 billion were completed without staff involvement, 44,000 commodity items were reclassified, and an extra $246,000 in revenue was identified.

Implementation
Indicative cost
High (€500k–€5M) — Enterprise-scale watsonx.ai/watsonx.data deployment plus LLM integration; likely a high-cost multi-year vendor engagement (exact contract value not published).
Time to results
Medium (1–3 years) — IT systems expanded from 9 (2021) to 37 (2025); AI classification system reached 80,000+ automated declarations in 2024.
Staffing & skills
IBM Expert Labs and IBM Client Engineering technical teams, State Customs Committee IT department (37 information systems by 2025)

Conditions for success

  • large historical declaration dataset for classification model training
  • integration across 33 government agencies' data systems

Common failure modes

  • no independently audited accuracy or error-rate assessment published

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