Two authorities, one union
This is the seventh paper in the African Revenue Intelligence Research Series, instantiating ARIM (Paper 1) for Tanzania. It introduces the series' first dual-jurisdiction structure (Mainland TRA vs Zanzibar ZRA), contrasting with Nigeria's federal-state PAYE fragmentation (Paper 4), while its intermediate EFDMS maturity and presumptive threshold bridge Rwanda's mature EBM model (Paper 6) and Ghana's threshold analysis (Paper 5).
On the Mainland, the TRA administers VAT at 18% and Corporate Income Tax at 30% under the Value Added Tax Act, CAP 148, as amended by the Finance Act, 2026 (in force from 1 July 2026). VAT registration is mandatory above TZS 200 million of turnover in twelve months, or TZS 100 million in six, and monthly returns fall due on the 20th of the following month.
Zanzibar is a genuinely separate jurisdiction rather than a devolved variant. The ZRA, renamed from the Zanzibar Revenue Board in 2023, applies a 15% standard rate, rising to 18% on banking, postal, telecommunications, insurance and digital services.
The 2026/2027 budget adds momentum: AI, big data and blockchain initiatives, integration of the TAUSI and IDRAS systems, a free mobile EFD-equivalent app for SMEs, a 12-month tax holiday for newly registered presumptive-regime taxpayers and a proposed rise in the digital services tax from 2% to 3%.
Data that is graded, never assumed
ARIM grades every source as essential, desirable, optional or potentially unavailable. TRA VAT and CIT returns, EFDMS transaction data and historical audit outcomes are essential. ZRA returns and TAUSI/IDRAS licensing data are desirable; mobile-app receipts are optional; and a formal TRA–ZRA data-sharing agreement is, today, potentially unavailable.
Tanzania’s EFDMS sits at intermediate maturity — richer than Nigeria’s transitional rollout, younger than Rwanda’s EBM. The cross-jurisdiction model therefore works from whatever Zanzibar-linked activity each taxpayer’s own Mainland filing declares, rather than assuming direct ZRA access.
Resolving who is who across a border
Deterministic matching on the TRA Taxpayer Identification Number links Mainland entities, and Fellegi–Sunter probabilistic matching uncovers shell networks and common control within the Mainland.
Matching a Mainland entity to a related Zanzibar-registered counterpart is a different matter. Without a formal TRA–ZRA agreement, ARIM flags it as not currently feasible with high confidence, and hypothesises materially lower match confidence across the border than within either jurisdiction. The gap is documented, not assumed away.
Three models for three risks
TZ1, cross-jurisdiction risk. It flags Mainland-only businesses that declare a disproportionate share of transactions as Zanzibar-based, and measures the VAT saved if that volume were shifted to the lower-rate jurisdiction. Zanzibar-side nodes in the taxpayer graph are inferred from Mainland declarations alone — a stated limitation.
TZ2, EFDMS anomaly detection. It compares claimed input VAT with EFDMS-verified input, holding any future mobile-app data in a separate, lower-confidence tier.
TZ3, presumptive-tax graduation. It scores proximity to the TZS 200 million threshold, while treating the new 12-month tax holiday as a legitimate reason for a post-registration pattern shift.
Six domains, one explainable score
ARIM scores each taxpayer as a domain-weighted sum across six behavioural domains. For Tanzania, each domain carries its own local variables, with the cross-jurisdiction feature as the most distinctive addition.
Tax behaviour
VAT-to-sales ratio deviation, CIT margin anomalies and proximity to the TZS 200 million threshold.
Transaction behaviour
EFDMS invoice-verification gaps, with a separately flagged tier for mobile-app data once deployed.
Network behaviour
Director and shareholder centrality plus cross-jurisdiction transaction-declaration density.
Sector behaviour
Sector-adjusted margin deviation, with mining and agriculture incentive-sector benchmarks.
Historical compliance
Prior TRA audit-adjustment count and magnitude.
External data
Declared Mainland–Zanzibar volume relative to business scale, and customs mirror-statistics gaps.
A flag is a signal, never a finding
Every case file carries domain-tagged reason codes. A cross-jurisdiction flag might read that 72% of declared transactions are attributed to Zanzibar with no corroborating ZRA registration or physical presence; a threshold case would instead note that the tax holiday is a plausible innocent explanation.
On a fully synthetic set of 30 taxpayers across eight sectors, the seeded cross-jurisdiction and EFD-suppression profiles ranked first and second, each driven by distinct domain evidence. The seeded TZS 195 million threshold case ranked eighth, deliberately, because proximity alone is a weaker signal than a confirmed transactional anomaly. Five hypothetical case studies, including a genuine two-sided tourism operator, test the false-positive safeguards.
Processing is governed by the Personal Data Protection Act, 2022. Any cross-jurisdiction flag is treated with heightened evidentiary caution, and the paper claims no real-world detection or revenue figures until the TRA validates the models on its own data.