Tax Tribunal: AI-driven innovation in administrative appeal work
Background
Tax appeal work is a specialized field in which cases are reviewed against a vast body of laws, administrative rulings, and court decisions. Each case has different facts and issues, so finding cases that look similar is not enough to complete a review. A separate step of comparing cases issue by issue is needed.
The same goes for writing decisions. The structure has to be consistent across organizing the facts, identifying the issues, reviewing the legal principles, and building the reasoning for the decision.
Existing general-purpose AI could be used to find and summarize documents. But it fell short at structured analysis grounded in the legal framework and at pinpointing the relationships between issues. That is why an analysis system reflecting the nature of tax appeal work was needed.
Platform overview
It is an AI system specialized for tax appeals, with the structure of the law built directly into its design. Laws and court rulings are broken down by chapter, section, article, and paragraph, and the relationships between provisions and issues are modeled as a structure. The key is that cases can be compared by what the issues are, not by how similar their text looks.
Answers are designed to present the relevant evidence, and analysis results are organized to match the structure of a written decision. A de-identification process for personal information was also built in to meet security requirements.
Key features
01 Case analysis AI
Analyzes appeal petitions and related materials, classifies the case type, identifies the main issues, and organizes everything around those issues.
- Shorter initial review through systematic organization of facts and issues
- More accurate similar case search based on structured analysis results
02 Similar case comparison AI
Searches for similar cases by issue and analyzes how closely cases resemble each other to support comparative review.
- Automatically links applicable laws and grounds for decisions to help organize the reasoning
- Less repetitive work in comparison and review, shortening overall review time
How it operates
Taking into account both the particular nature of tax appeal work and the operational requirements of public institutions, the design combines analysis based on legal structure with a security-first approach.
- Laws broken down and systematically organized by chapter, section, article, and paragraph
- Relationships between case issues modeled as a structure for accurate analysis
- Answers generated with search results as their evidence
- An AI model specialized for tax work, applied for a better fit with real practice
- De-identification of personal information and security controls in place
It was designed to run reliably even on an internal network separated from the outside.