Everything advisors, in-house tax teams and tax authorities ask.

exPErtus is a B2B SaaS platform that automates the legal analysis of Permanent Establishment (PE) risks for multinational enterprises and the firms that advise them. It delivers complete, source-referenced legal memoranda within minutes — a process that traditionally takes hours or days of senior advisory time.

A Permanent Establishment — "PE" — is a foundational concept in international taxation. In simple terms, a PE defines when a company operating across borders creates a taxable presence in another country. Once that happens, the consequences are real: the company must register, file corporate tax returns, attribute profits to that local presence, and potentially deal with payroll tax, social security and VAT obligations.

A concrete example illustrates the issue. Imagine a Belgian engineering company that wins a project in the Netherlands, requiring a team of engineers to work on-site in Rotterdam for several months. The same company has rented a small warehouse in Germany to store equipment for clients across Europe. And one of its senior managers, originally Belgian, now works permanently from a home office in Cyprus.

Each of those situations — the project in the Netherlands, the warehouse in Germany, the remote employee in Cyprus — could potentially create a PE in those countries. If they do, the Belgian company (and its employees) become a non-resident taxpayer there and must comply with local tax rules. The financial and administrative consequences can be enormous, and they often surface only years later during a tax audit, when corrective action is no longer possible.

What makes PE analyses particularly complex is that every case requires a fact-specific assessment. Whether a PE exists depends on the nature of the activities, their duration, the function of the people involved, whether they relate to core or merely preparatory activities, the extent to which people can negotiate or sign contracts, and more. It is not a yes/no checklist — it requires structured legal reasoning across multiple layers of rules and sources, starting from the applicable tax treaty and the domestic tax rules. PE advice is therefore not a commodity; companies that want to monitor PE exposure proactively typically rely on tax advisors all over the world, which makes it very costly. That is the problem exPErtus solves.

The good news — and the reason exPErtus can scale globally — is that the PE concept is remarkably consistent worldwide. The definition between Belgium and the Netherlands is +95% the same as between Germany and Austria, or Japan and Australia. Most countries base their tax treaties on the OECD Model Tax Convention, which contains a detailed definition and — through the OECD Commentary — detailed guidance on the PE concept. Local variations exist, but the core analytical structure is universal.

exPErtus delivers a full legal memorandum — the same output a top-tier tax advisor would produce. This includes a structured analysis of the relevant facts and theoretical framework, condition-by-condition reasoning based on the applicable treaty provisions, a full audit trail with explicit source references (OECD Commentary, treaty articles, case law, doctrine, domestic legislation) for each conclusion, and a clear, defensible recommendation.

The same underlying analysis is available in multiple formats: a comprehensive legal memorandum for the file, a concise written advice for client communication, a ruling request for the tax authority, or a defence file for an audit context.

Three target segments, with one common denominator — a need for fast, consistent and legally robust PE assessments.

• International tax advisory and law firms handling high volumes of PE questions under margin pressure.
• Large multinational enterprises with in-house tax teams trying to solve PE assessments internally, with senior tax people spending too much time on these issues.
• Tax authorities seeking to increase audit efficiency and broaden their taxable base.

Two interaction modes, designed to fit seamlessly into existing practice.

Structured chat interface. Mirrors how advisors already gather facts. The tool identifies what information is needed, helps gather it through targeted questions, and produces the analysis automatically.

Email-based interaction. The advisor initiates the process by forwarding (or composing) a client's email to the exPErtus platform. The tool processes the facts, identifies what's missing, and sends targeted follow-up questions directly to the client on behalf of the advisor. The client's responses are captured automatically. If further clarification is needed, the tool follows up — again by email — until the fact set is complete. Once all inputs are available, exPErtus runs the full PE analysis and delivers the final memo to the advisor for review. The advisor remains in control throughout: the memo is delivered to the advisor (not the client) and is only shared with the client after the advisor has reviewed and approved it. Nothing fundamentally changes in the workflow — only the speed and consistency.

exPErtus is built around a massive proprietary reasoning structure that mirrors how experienced PE specialists actually think — condition by condition, factoring in the interaction between conditions, with the relevant authoritative sources. This framework was designed first, independently of any AI model, based on decades of practice experience. AI is then used as an execution layer, not as a decision-maker. A controlled language model follows predefined reasoning paths, applies mapped authoritative sources, and articulates conclusions consistently.

AI plays a critical but controlled role. It executes a reasoning structure designed and validated by PE experts. The AI does not "decide" the legal analysis — it follows explicit, incredibly detailed and predefined reasoning paths. It applies specific treaty provisions, OECD Commentary, case law and doctrine to facts, exactly as instructed.

Think of it as a PhD-level skilled senior associate following a senior partner's exact methodology — but always consistently, fast, and at scale.

The architecture is specifically designed to prevent hallucinations. Unlike generic AI tools that generate responses from probabilistic language patterns, exPErtus constrains the AI within a proprietary reasoning framework. Every analytical step is mapped to specific authoritative sources. The AI cannot deviate from these paths or invent conclusions. Each output includes explicit source references that can be independently verified.

The system evolves through an anonymised, automated feedback process combined with controlled expert oversight. Rather than learning directly from individual outputs, feedback is aggregated and anonymised before being analysed. When patterns of improvement or errors are identified, they are reviewed and validated by domain experts. Only after expert approval is the insight incorporated back into the system. This keeps learning accurate and legally compliant, and avoids biases or unintended behaviours.

Yes. Because the reasoning structure is independent of any specific AI model, exPErtus can adapt to improvements in underlying language models without rebuilding the core engine. If a more capable model becomes available, it can be integrated as the execution layer while the proprietary reasoning architecture remains unchanged. This protects against model obsolescence and allows continuous improvement.

All current AI tools for legal and tax work share the same architecture, regardless of how they market themselves. From general-purpose models (ChatGPT, Claude) to specialised legal-AI platforms to tax-specific tools, every system follows the same four-step pattern:

1. Retrieval — search a curated legal corpus for documents relevant to the query.
2. LLM — a frontier model (GPT-5, Claude Opus 4.6, Gemini 2.5) reads the retrieved documents.
3. Synthesis — the LLM generates an answer, citing the documents it received.
4. Output — the user receives a draft.

This is "retrieval-augmented generation" (RAG). It is fast, scalable and works well for research-style tasks. But for high-stakes legal reasoning, three structural weaknesses are well documented.

The reasoning path is determined by the LLM, per query. Every time you ask a PE question, the LLM independently chooses what to retrieve, what to weigh and how to synthesise. The same case asked twice can produce different answers — even at maximum determinism settings — as documented in Magesh et al. (2025), Journal of Empirical Legal Studies.

Sources can be hallucinated or missed. If retrieval misses a relevant treaty paragraph, the LLM either fabricates one or works without it. Magesh et al. (2025), Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools, Journal of Empirical Legal Studies, found leading legal RAG platforms hallucinate in 17–33% of test queries — and that is the better end of the market, well below ChatGPT's documented failures (the Mata v. Avianca case being the canonical example, with sanctions imposed on a New York lawyer who filed a brief containing fabricated case citations).

There is no audit trail of the reasoning itself. RAG tools document what sources were retrieved, but not why the LLM took the analytical path it did. For a tax authority asking "how did you reach this conclusion?", "the AI synthesised this from these sources" is not a satisfying answer.

exPErtus inverts the architecture. The reasoning is not what the LLM does — the reasoning is the proprietary decision framework, authored by senior PE specialists over decades of practice. Every PE question has a predefined path through that framework. Each node carries an explicit legal question (e.g. "Is the home office at the disposal of the enterprise under the post-2017 Commentary?") and the exact authoritative sources that apply at that point — treaty article, OECD Commentary paragraph, marginal notes in authoritative doctrine, relevant case law, local administrative guidance. The next node depends deterministically on the answer to the current one. The path through the tree is the legal reasoning.

The LLM only executes that path. It does not decide what to retrieve, what is interpretive, or what to cite. It walks the predefined nodes, applies the predefined sources, and articulates the conclusion in clean professional language.

The consequences:

• Deterministic. Same case, same answer, every time.
• Bounded sources. Nothing is dynamically retrieved; nothing can be invented. Every citation in an exPErtus memo is in the framework, has been verified by senior specialists, and applies to the specific question being analysed.
• Audit-traceable. Every conclusion ties back to a specific node and that node's predefined sources. The reasoning artefact is the framework — not an opaque LLM trace.

This is what we mean by "two-tier reasoning engine" and "we started with the legal reasoning, the AI only executes it." It is a fundamentally different architecture, not a marketing distinction.

Beyond the LLM-based reasoning tools covered above, the broader TaxTech landscape includes other AI-powered categories that solve adjacent — but different — problems:

• Tax governance & compliance platforms. Track obligations, flag risks and monitor deadlines. Useful for compliance hygiene — they do not perform in-depth legal PE analysis.
• Tax preparation & compliance tools. Automate transactional calculations and filing. Strong on numbers, no advisory-level analysis.

exPErtus is the only tax platform that starts from a proprietary reasoning framework, interprets and applies authoritative sources on a consistent basis, and delivers a complete top-notch tax memorandum from a complete set of facts. Other tools tell you what to file, when to file it or where the risks are. exPErtus tells you whether — and why — there is a PE in the first place, with the audit trail to defend it.

No huge consulting firm currently offers a standalone PE analysis product, and there are strong structural reasons why this is unlikely to change.

Large consulting and law firms are investing heavily in AI, but their approach is fundamentally different. They are building broad, horizontal AI tools to cover all aspects of their legal and tax practice at once — general-purpose research assistants, document review, workflow automation. These are data-driven tools: LLM layers on top of internal knowledge bases and external databases that help professionals find information faster or draft advices that are slightly better than generic AI, but not consistent, authoritative, complete and fully trustworthy. That is research acceleration, not a reasoning architecture.

Building a deep, niche-specific reasoning engine — one that mirrors how a PhD-level senior associate actually thinks, condition by condition, source by source — is an entirely different undertaking. It requires a dedicated, multi-year investment to design, validate and maintain a structured decision framework for one sub-domain of tax law. For a Big Four firm with hundreds of practice areas competing for innovation budget, that level of specialised investment in a single niche is extremely difficult to justify internally. Auditor-independence rules also prohibit them from relying on a tax-exposure assessment tool developed by their own firm.

The far more logical outcome — and the one we are actively pursuing — is that large consulting firms become exPErtus' clients. For the cost of an annual licence fee, they gain access to a tool that materially improves PE practice economics, without cannibalising their client relationships. They remain the trusted advisor; exPErtus simply makes the work faster, more consistent and dramatically more profitable.

The exPErtus moat operates on three levels:

• A massive proprietary reasoning structure. A complex logical architecture built from decades of PE practice. Not starting from a dataset (as other TaxTech tools do), but a structured decision framework.
• Source mapping & interpretive layer. Each reasoning node is linked to specific sources and includes interpretive logic for applying those sources to facts. This is where generic AI tools fail.
• Lockstep expertise requirement. Building a tool like exPErtus requires deep PE expertise, decades of practical judgement and technical implementation working together. IT developers alone cannot build it — buying the legal knowledge would cost a fortune. Tax lawyers alone cannot manage either: PE expertise is fragmented, mostly sitting with academics (without practical experience) or Big Four (restricted from building such a tool by independence rules). That combination is precisely why no one has tried this before, despite the obvious market opportunity.

No. Seeing the output does not reveal the architecture. The final deliverable is a complete top-notch legal memorandum, but the value lies in the reasoning engine underneath: hundreds of decision nodes, each mapped to specific authoritative sources, with interpretive logic calibrated through decades of practice. The reasoning engine forms the unique IP of exPErtus and is protected to the maximum extent.

Trying to reverse-engineer the reasoning architecture from the output would be like trying to derive the Coca-Cola formula by drinking a glass — no matter how carefully you taste it, the recipe stays hidden.

The commercial launch is foreseen for Q4 2026.

In the meantime, the MVP is ready end of May 2026 and pilot testing with committed partners opens in June 2026. Ten organisations have committed to pilot testing, and several Letters of Intent have already been received. A limited number of pilot slots remain — see the waitlist on the home page.

exPErtus' first module covers PE assessment. From the commercial launch onwards, we extend the same reasoning-first architecture to adjacent areas: profit allocation, VAT/GST presence and cross-border workforce taxation. The vision is to become the global standard for assessing and managing taxable presence risk — a decision infrastructure for cross-border tax across all relevant treaty jurisdictions.

Yes. Every output is source-referenced and follows orthodox PE reasoning methodology. Because of the audit trail, users can trace each conclusion to the specific treaty provision, OECD Commentary paragraph, case law or doctrine supporting it. The output is suitable for internal decision-making, sharing with external advisors, and defending positions during a tax audit.

No — it transforms what they can do. For advisory firms, exPErtus dramatically reduces lead time and cost per analysis, allowing them to serve more clients at materially better margins. The trusted advisor relationship is preserved; the engine simply makes the work faster, more consistent and more profitable.

The reasoning engine accommodates legal change through a structured maintenance process overseen by the expert team. Because the architecture is modular and source-mapped, updates can be implemented precisely without disrupting the broader system.

This is further strengthened by exPErtus' structural collaboration with a world's leading independent provider of cross-border tax expertise. Through this partnership, treaty changes, new OECD Commentary updates, relevant case law and legislative developments across jurisdictions are systematically channelled and pre-processed — ensuring that the reasoning engine always operates on the most current and authoritative sources.

Data security and regulatory compliance are foundational to the exPErtus architecture. Client data is processed in a controlled, closed environment hosted within the European Union. The AI model does not learn from individual client inputs — training data and assessment data are strictly separated, ensuring no client-specific information is ever used to train or refine the underlying model.

exPErtus is designed from the ground up to comply with the General Data Protection Regulation (GDPR), including data minimisation, purpose limitation and storage limitation. Only data strictly necessary for the PE assessment is collected and processed; client data is not retained beyond the period required for the engagement. exPErtus is built in alignment with the EU Artificial Intelligence Act (AI Act), which introduces a risk-based regulatory framework for AI systems. As a tool that supports — but does not replace — professional legal decision-making, exPErtus operates with full transparency regarding its reasoning process, source references and the role of human oversight, consistent with the AI Act's transparency and human-in-the-loop requirements.

exPErtus is pursuing ISO 27001 certification (information security management) to provide enterprise clients with independently verified assurance regarding data handling, access controls and incident response procedures. Detailed security specifications — including penetration testing reports, data processing agreements and sub-processor documentation — will be available on request.