ZRA AI Unit — Zambia Revenue Authority's Machine-Learning Customs Risk Model

Since around 2023, the Zambia Revenue Authority (ZRA) has operated an in-house 'AI Unit' tasked with modernising tax and customs administration through machine learning. In partnership with UC Berkeley's Center for Effective Global Action (CEGA), the unit adapted a Random Forest model — previously piloted on Paraguayan customs data — to flag high-risk import declarations for physical inspection.

Address an estimated 47-56% customs tax gap across roughly half a million commercial shipments processed annually, by targeting inspections toward the highest-risk declarations rather than relying on blanket or purely manual checks.

The customs risk-scoring pilot began around November 2025 and, according to CEGA, is being evaluated through a randomised design. The AI Unit also runs adjacent tools, including a taxpayer chatbot, an HS-code lookup assistant, and automated compliance 'nudges.'

Quantitative results on detection accuracy or revenue recovered specifically attributable to the Random Forest model had not yet been published as of the pilot's start. ZRA and Zambia's Ministry of Technology and Science have separately stated that AI-driven modernisation across the revenue authority's operations contributed to a 60% increase in revenue collection over two years, but that figure describes the broader AI programme rather than this specific model.

The 60% revenue-growth figure has not been independently audited and should be read as an institutional claim about the wider AI programme, not a verified causal result of the customs risk model specifically; randomised pilot results for the model itself are still pending.