Global Tax Analyzer is EY's tax AI platform tuned to Korean tax law. One question retrieves statutes, authoritative interpretations, tribunal decisions and court cases in real time, answers with the evidence, and turns it into ready-to-use documents — opinion letters and memos.
Tax answers you cannot trace are answers you cannot use. Global Tax Analyzer checks statutory and case citations against official texts, and clearly flags any that remain unresolved — then connects the evidence to ready-to-use documents and EY professional review.
Every question retrieves statutes, authoritative interpretations, tribunal decisions and court cases directly from official data — new amendments and cases apply immediately, with no retraining.
Recognized citations to statutes, cases, tribunal decisions and rulings are routed to source-specific checks for existence, provision structure, decision dates and quoted excerpts. Anything outside automatic lookup is clearly marked as such.
Download opinion letters, memos and response statements in firm templates — and matters that call for judgment are handed on to your EY tax professional.
This isn't a general chatbot with tax bolted on. Global Tax Analyzer was designed and built by practicing EY tax professionals — it finds the actual statutes and cases, routes recognized citations through source-specific checks and makes the verification scope explicit.
Tax practice and verification principles are implemented in code. Core determinations — rates, valuation, risk — are computed by deterministic engines; the AI writes the explanation and the document on top. Plausible-but-wrong answers have nowhere to hide.
We cross-check every statute and case citation against the official source for two things: whether it actually exists and whether the quoted wording matches the original. We check down to the paragraph and subparagraph — a wrong one is corrected to the right one, and a reference we cannot confirm is never left in the answer; the checking is never handed back to you as homework. A quoted excerpt that differs from the source is fixed to the actual wording or labeled as a gist, so an unverified sentence never looks like a verbatim quote. Decision and judgment dates are also checked against the official source. Citations that remain unconfirmed are not promoted as controlling support; they are removed or carry an explicit status according to the coverage available for their source type. Answers never use an internal search tool or storage path as a proxy for reliability; status is described only through the original text, identifier, applicable date, and available text scope. The soundness of interpretation and application, however, is flagged as an expert-review matter — existence checking is never dressed up as a guarantee of correctness.
Enter a tax year or transaction date and the analysis runs under the law in force then — exactly what claims for correction, audit defense and prior-year reviews require.
Tax materials are processed within your organization's boundary, and the platform can be deployed in a customer-owned cloud or on-premises when required.
Explicit document numbers are checked against the official text first. One investigator then chooses between live official tools and the supplemental corpus, preserves the full reasoning and order sections of long decisions, and admits only selected evidence to the Evidence Ledger before answer, citation and point-in-time verification. If decisive evidence is still missing, it runs one bounded supplement round. High-risk matters are marked for professional review.
Turn research into opinion letters, memos, emails and response statements — export to PDF · Word · Hangul (HWP), or into your own corporate template.
A table shows which sentence of which source each judgment rests on. Every basis leads with its type ( statute · interpretation · precedent) and a plain-language read, so you can follow the reasoning without any tax expertise. Tap a number to open the source.
Attach contracts or internal memos in PDF, Word, HWP, HWPX or text and the review reflects your facts. Uploads are used only for that review, automatically deleted after one week, and never used for model training.
Admin-curated hashtags, coverage teams and credentials are matched against the question and the verified answer topic to surface the Head of Tax, the relevant division leader and 2 to 4 partners. No-match cases keep the established leadership list; simple statute and case-number lookups omit routing.
Interface and deliverables in both Korean and English — ready for global reporting and cross-border work.
Beyond the general research chat, dedicated agents are built for recurring tax workflows. Agent availability and feature scope are provided selectively by contract and plan, and some agents are for EY internal use only — certain features such as diagnostics provide summary-level results, with in-depth review delivered through EY professional engagements.
National and local tax legislation and the managed corpus remain available. When an official MCP outage is detected, matching corpus texts are injected automatically so answers can continue, with incomplete currency and official rechecking disclosed.
Citations that can't be auto-verified are never silently deleted — they're flagged "verify directly." We never conclude something "doesn't exist" just because our search didn't find it — no official database has complete case-law or ruling coverage, so rather than deleting a real citation and breaking trust, we mark it "outside automatic lookup" and point you to the source. A confidence level is stated too.
Access control, concurrent-session management and audit logging are built in; models range from the in-house engine to frontier LLMs by contract.
The order in which evidence is found is part of accuracy. When a case or document number is supplied, the platform resolves the official source before any model judgment. A single investigation flow then considers live official sources and the managed tax corpus together, admitting only evidence with recorded provenance and verification status. Conclusions follow the law applicable at the transaction date → authority hierarchy → recency among evidence of the same authority and issue. Recognized citations and temporal applicability are checked again after drafting, and low-relevance material is never retained just to fill a quota.
Just ask. From "repudiation of wrongful calculation on a low-price transfer to a related party" to contract review and unlisted-stock valuation — one line of question comes back as evidence and a document.
Tax law changes often, and a court decision interprets the law as it stood at the time. So before ranking authorities, the platform first confirms which statute version applies to your taxable year and whether it has since been amended.
Live official lookup and the managed tax corpus are not searched in isolation and merely appended. The platform selects what the issue and authority mix require, records official cross-check and text availability, and only then passes evidence to answer generation. Structured DART company analysis remains a separate menu outside this research flow.
We verify which statute version a precedent or ruling interpreted; if the law has since changed, it is not used as direct authority for today's conclusion. Older materials are kept — clearly marked as prior doctrine and amendment background — rather than deleted.
Court reasoning and current administrative practice can diverge. We present both sides rather than one, and explain the practical burden this can create at filing or appeal.
Sources that cannot be verified carry their verification status and limits, stated as such. Where useful, we point to official advance-ruling procedures (NTS advance answers, written inquiries) or expert review — reducing uncertainty before you act.
Usage is managed transparently in credits. The in-house tax engine runs without limits; premium models are chosen only when you need more power. Scale and rates are configured to fit each contract.
Research and review on EY's dedicated GTA engine come with a generous allowance included in your plan. The scope of inclusion is configured per contract.
For matters that call for deeper reasoning, you can select a frontier model. Only premium usage is metered in credits — you pay for what you use.
Administrators assign seats, credit limits, and model access per organization and member, with usage visible transparently on a dashboard.
Scope, model mix, and rates are designed around the size and needs of your division or entity. Enterprise agreements and dedicated-model integration are supported.
Cost is governed through concurrency, queueing, and limits — we never shorten output to save cost. Even long review memos are completed end to end.
Uploaded materials and conversations are processed within your organization's boundary and are not used to train models. Dedicated-cloud or on-premises isolated deployment is available.
Spacing and hyphen variants resolve to the same identifier. Decision and judgment dates are compared with official text, and low-relevance material is not retained to fill a quota.
Annexes, attached forms, and rate tables are checked with the governing text. Restriction of Special Taxation Act trends (integrated investment, national strategic technology, new-growth R&D, employment) are provided in a separate workbench.
Company screening conditions such as semiconductor sector × operating margin are not mixed into the free-form investigator; they are decided in a dedicated DART menu against official disclosure fields.
Original attachments are not retained as a long-term archive. Temporary extracted material is automatically deleted after one week.