Leticia Arthur
AI & Data Science MUSTSmartKodiAI
Problem Statement
Tanzania aims to transform its economy toward a USD 1 trillion economy under Vision 2050 and FYDP IV. To support this growth, the Ministry of Finance targets domestic tax revenue of 13.3% of GDP, while TRA has set a revenue target of TZS 41.830 trillion for the current fiscal year. However, revenue mobilization faces three key challenges: limited visibility of economic activity in the informal sector, revenue leakage through practices such as invoice splitting, under-reporting, undervaluation, and coordinated transactions, and heavy reliance on retrospective manual audits. Although systems such as EFRS and IDRAS have strengthened digital revenue administration, there remains an opportunity for an AI-based analytical layer to identify unusual transaction and business patterns.
Proposed Solution
SmartKodiAI is an AI-powered tax revenue intelligence and anomaly-detection service designed to operate above existing government revenue systems rather than replace them. It analyzes authorized transactional and business metadata to identify unusual patterns, transaction networks, and behavioral signals that may warrant further compliance investigation. For example, a group of business TINs may show repeated circular transactions, unusually fragmented invoices, or transaction values that are materially inconsistent with observed patterns in a relevant sector. SmartKodiAI can rank the pattern as anomalous, generate a risk signal, and provide an explanation of the main factors contributing to the flag. The authorized auditor then decides whether further investigation is appropriate. The service therefore acts as an intelligence layer connecting existing revenue data with analytical insight: existing systems continue to perform their operational functions, while SmartKodiAI adds anomaly detection, risk prioritization, network-pattern analysis, and explainable AI.
Innovation
SmartKodiAI's main innovation is not the creation of another tax administration database. Its innovation is the use of AI as an intelligence layer over authorized revenue and government data sources. Instead of waiting for manual review after suspicious activity has already been identified, the system continuously analyzes patterns and prioritizes cases that are unusual relative to learned behavior.
Expected Impact
Success will be demonstrated when SmartKodiAI can reliably detect unusual transaction patterns while maintaining an acceptable false-positive rate. In the long term, success will mean better audit prioritization, less time spent reviewing low-risk transactions, improved identification of high-risk transaction networks, and stronger evidence for auditors. A future target is a 60% reduction in manual audit processing cycles, to be validated through a controlled pilot. Success will also include explainable AI, human oversight, privacy protection, access controls, and continuous monitoring of model performance.