The Tribunal does not read your model. It reads your reasoning.
Section 56 of the Tax Procedures Act places the burden of proof on the taxpayer. That is the law. What it means in practice is that the Commissioner's assessment must, on its face, carry the reasoning the Tax Appeals Tribunal will read against the taxpayer's evidence. The assessments that fail at TAT are not failing on the law. They are failing because the file does not show how the Commissioner thought.
There is a meeting that happens at the Kenya Revenue Authority's headquarters in Times Tower at the end of every quarter. It is the meeting where the senior officers in the Domestic Taxes Department review the Tax Appeals Tribunal docket. The numbers behind the meeting are the ones the Commissioner General reads to the Cabinet Secretary for the National Treasury: how much is in dispute, how much the authority has won, how much it has lost, how much is now sitting at the High Court on appeal. The question underneath the numbers is harder. The assessments that lost — why did they lose.
The published TAT decisions over the past eighteen months tell a consistent story. In TAT E1149 of 2024, Premier Credit Ltd appealed a KES 30 million corporate income tax assessment over disallowed bad debt deductions. In TAT E134 of 2025, Manchester Outfitters Ltd appealed a KES 53 million assessment for failure to file. The Del Monte Kenya decision in early 2026 upheld the KRA's position because the taxpayer 'failed to discharge its statutory burden of proof having relied on mere averments and correspondence rather than registry records, source documents, and verifiable financial data.' The pattern across the docket is the same. The Tribunal is looking at what each side has put in front of it and deciding on the evidence.
Section 56(1) of the Tax Procedures Act places that burden on the taxpayer. In the High Court's January 2022 decision in the Pearl Industries case, the bench affirmed that both the Tax Appeals Tribunal Act and the Tax Procedures Act 'impose the burden of proof on the taxpayer to prove that an assessment is excessive or incorrect.' The taxpayer must produce evidence to dislodge the Commissioner's findings. That law is now itself under constitutional challenge — Petition E002 of 2026 in the Milimani Constitutional Division is asking whether Section 56 tilts the field too far against the citizen.
But the burden of proof on the taxpayer does not mean the Commissioner's reasoning is invisible. It means the Commissioner's reasoning is the floor the taxpayer must dislodge. When the assessment is well-reasoned, the taxpayer must produce strong evidence to overturn it. When the assessment is thinly reasoned, the taxpayer's evidence does not need to be very strong to introduce doubt. The Tribunal then has to choose between two pieces of inadequate work.
The burden of proof is on the taxpayer. The burden of reasoning is on the Commissioner. Confuse the two and you lose at TAT.
The pressure on assessment quality is rising on every front. KRA's Income & Expense Validation Engine, live from 1 January 2026, will cross-check 2025 returns against eTIMS, withholding tax filings, customs imports and PAYE. The volume of additional assessments — where declared income or expenses do not reconcile with the engine's data — will rise. The Intelligence Analysis Tool, scheduled for full rollout in June 2026, will widen the surface further by bringing in inter-agency and open-source data. The Draft Finance Bill 2026 proposes giving the Commissioner statutory authority to determine liability from secondary data in cases of suspected tax avoidance, moving the system structurally away from voluntary self-assessment.
FIRS in Nigeria is on a parallel path. The Federal Inland Revenue Service's accreditation of PwC Nigeria as a system integrator for the FIRSMBS e-invoicing platform in December 2025 is part of a structural shift to transaction-level reporting. Sections 25 and 26 of the FIRS Establishment Act 2007 already empower the deployment of technology to automate assessments. The Nigeria Tax Reform Acts 2025 went further, giving FIRS broader audit authority. Every additional assessment FIRS raises must, sooner or later, defend itself before the Tax Appeal Tribunal — and on appeal to the Federal High Court.
Where each sits.
Akki sits between the authority's tax administration system (iTax in Kenya, FIRSMBS in Nigeria, the RRA platform in Rwanda, eFiling in South Africa) and the validation engines, third-party data feeds, and AI tools the authority deploys. Every assessment file the authority builds passes through Akki. The taxpayer's filings, the third-party data points, the validation engine outputs, the officer's analytical work, the recommended assessment — all logged, all retrievable, all attached to the file in a form the Commissioner's legal team can produce at TAT.
Solva is the layer that produces the reasoning the Tribunal will read. When an officer or a model identifies a discrepancy and proposes an additional assessment, Solva structures the reasoning through five stages. It restates the question — is the declared position inconsistent with the data the authority holds. It surfaces what is known about the taxpayer's filings, what the third-party data shows, and what would reconcile the two. It triangulates between the taxpayer's history, the sector norms, and the authority's prior treatment of similar files. It produces a recommended assessment with the conditions explicit. When the evidence does not support an assessment that would survive Tribunal review, Solva refuses to recommend and names the gap rather than producing a thin file that will lose. The refusal behaviour is the feature the Commissioner most cares about. An authority that pursues every discrepancy regardless of evidence quality wins less at TAT and creates the public reaction the IAT clarification was issued to address. An authority that pursues only the assessments its own reasoning supports wins more, loses less, and absorbs less public backlash.
SyniSense governs the data layer specifically for citizen data. The third-party data the authority brings into an assessment — bank records, mobile money flows, employer filings, customs declarations, NIBSS BVN data, eTIMS invoices — is sensitive under the Data Protection Act 2019 in Kenya, the NDPA 2023 in Nigeria, Law n° 058/2021 in Rwanda, and POPIA in South Africa. SyniSense ensures that the reasoning the AI models do happens against data anonymised at the boundary, with re-identification only inside the authority's environment.
The Domestic Taxes Department's quarterly TAT review changes. The conversation is not about which assessments lost. It is about which assessments should not have been raised in the first place — and the file shows why each one that proceeded was worth pursuing.
The Commissioner's legal team's preparation changes. The team is not reconstructing the reasoning from emails and worksheets. They are reading the assessment's own reasoning trail and producing the statement of facts at the Tribunal from a record that is already complete.
The taxpayer's experience changes. The objection notice the taxpayer receives is specific to their file. The reasons given are the actual reasons. The objection conversation, often resolved through Alternative Dispute Resolution with its 90-day window, happens on the evidence.
And the conversation about Section 56 changes too. The constitutional question about whether the burden of proof on the taxpayer is fair becomes a softer question when the Commissioner's assessments routinely carry their own reasoning. The system feels less arbitrary because, on each individual file, it is not.