When your knowledge is everything, losing hours finding it is losing the business
US Tax Service is the leading US tax advisory practice for Americans abroad in Switzerland. Their team of CPAs and IRS Enrolled Agents guides clients — from individual expatriates to multinational corporations — through the most complex intersections of US federal tax law, Swiss cantonal rules, and international treaty obligations. Every engagement touches FATCA, FBAR, Form 5471, FIRPTA, the US-Swiss Estate Tax Treaty, streamlined filing procedures, and a dense web of IRS guidance that changes regularly across multiple jurisdictions.
As the firm grew — expanding its reach from Zürich into Frankfurt, Amsterdam, London, Seoul, and Dubai — so did the volume and complexity of the knowledge it needed to manage. Regulatory publications, client case archives, internal advisory memos, IRS announcements, OECD treaty documents, cantonal tax guidance, and prior engagement records had accumulated across a fragmented ecosystem of shared drives, a legacy document management system, and email archives.
The result was a daily tax on advisor productivity that had become impossible to ignore.
Regulatory guidance, case precedents, and internal memos lived in siloed systems with no unified search layer. Finding anything cross-jurisdictional required knowing which system to look in first.
Existing search tools matched terms but not meaning. An advisor searching for “FBAR penalty mitigation” would retrieve documents with neither concept correctly contextualized for their client's circumstances.
Before providing substantive guidance, advisors spent an average of six hours per complex engagement searching, reading, and cross-referencing documents that should have been seconds away.
As the firm grew across new geographies, expertise developed in Zürich was not easily accessible to teams in Frankfurt or London. Senior advisor knowledge was siloed in individual heads and inboxes, not a shared system.
CEO Ian Young recognized this not as a search problem but as a knowledge architecture problem — and one that would compound with every new market the firm entered. US Tax Service engaged Sphere Partners to design and build a solution from the ground up.
“Tax advisory is a knowledge-intensive business built on precision. Before Sphere, our advisors were spending upwards of six hours per engagement just searching for documents that should have been seconds away. That is not a technology gap — it is a competitive liability.”
A production-grade enterprise RAG architecture built for regulated professional services
Sphere Partners began with a structured discovery process rather than immediately proposing a technology stack. Understanding the specific failure modes of US Tax Services AG's existing retrieval workflows — and the compliance sensitivity of the data involved — was a prerequisite to designing a system that would actually work in a regulated professional services environment.
Knowledge source mapping & data governance architecture
Sphere's team inventoried every document repository in the firm's ecosystem: IRS publications and revenue rulings, OECD treaty documentation, Swiss cantonal guidance, internal advisory memos, client case archives, engagement letters, and research from prior matters. Critically, they mapped the sensitivity and access requirements for each source type — establishing a role-based access control model from the architecture stage, not as an afterthought. This ensured advisors could only retrieve content within their authorization scope, and that client-specific records never surfaced in cross-advisor queries.
Domain-optimized document ingestion pipeline
Tax and legal documents are structurally unlike the documents most RAG systems are trained on. They contain dense cross-references, defined terms, numbered subsections, and jurisdiction-specific language that generic chunking strategies handle poorly. Sphere built a custom document ingestion pipeline with chunking logic optimized for long-form regulatory and legal text — preserving semantic units like numbered provisions and defined-term contexts rather than splitting them arbitrarily at character count boundaries. Metadata was extracted and attached to every chunk: document type, jurisdiction, applicable form or provision, publication date, and recency status.
Jurisdiction-aware semantic retrieval with hybrid search
The retrieval layer was built on a hybrid vector and keyword search architecture— combining dense embeddings for semantic similarity with sparse retrieval for exact regulatory citations (e.g., “IRC §911” or “Form 5471 Category 5”). A jurisdiction-aware metadata filterwas implemented so that queries automatically scoped to the applicable regulatory environment — US federal, Swiss cantonal, German, Dutch, or UAE — based on the client context supplied by the advisor. This eliminated the false positives that plagued the firm's previous search: retrieving Swiss tax guidance in response to a US federal FBAR query, or surfacing outdated IRS rulings that had been superseded.
RAG evaluation benchmarking at every milestone
Before deploying any interface, Sphere ran structured RAG evaluation benchmarksacross four dimensions: retrieval precision, retrieval recall, answer faithfulness (whether the generated response was grounded in the retrieved documents), and answer relevancy (whether the response actually addressed the advisor's query). Sphere's position was that deploying a RAG system without rigorous pre-production evaluation is the single most common failure mode in enterprise AI implementations — and they refused to skip it. Each milestone had a quantitative pass/fail threshold before the build progressed.
Audit trail & compliance layer
Every query, retrieval event, and generated response is logged in a tamper-evident audit trailthat records the source documents retrieved, the chunks cited, and the model's synthesis. This gave US Tax Services AG's managing directors full visibility into how the AI system was being used, and provided an evidentiary record for any engagement where the AI-assisted research might later be subject to professional review. In a regulated advisory environment, this is not optional — and Sphere built it into the core architecture rather than bolting it on post-launch.
Production-ready in five weeks. Measurably transformative from day one.
The impact of reducing per-engagement research time by 97% compounds rapidly across a professional services firm. For US Tax Service, it means advisors can take on more engagements without adding headcount. It means junior advisors can access the same depth of institutional knowledge as senior partners. And it means the firm's expansion into Frankfurt, Amsterdam, London, Seoul, and Dubai can be powered by a knowledge base that travels with the business — not a base of expertise that stays in Zürich.
“Sphere deployed a production-ready RAG pipeline in five weeks. Document retrieval accuracy improved by 66% compared to our previous keyword-based search. Most strikingly, the time our advisors spend on document research dropped from an average of six hours per engagement to under seven minutes. Sphere understood that responsible AI means the governance layer is designed in — not added later. For any professional services firm drowning in document complexity, Sphere is the partner I would call first.”
Enterprise RAG for regulated professional services: a different problem than most vendors solve
Most enterprise RAG implementations are built for general-purpose document Q&A — a simple vector search over a document repository with an LLM generating answers. This works well for internal wikis, HR policy documents, and product documentation. It fails for tax and legal professional services for several reasons.
Regulatory documents are structurally adversarial to generic chunking. An IRS revenue ruling or a bilateral tax treaty contains extensive cross-references, hierarchical subsections, and defined terms whose meaning only resolves in context. Generic character-count chunking splits these structures arbitrarily, producing retrieval chunks that are semantically incomplete and faithfulness-degrading when used as RAG context.
Jurisdiction matters as much as topic.“Penalty abatement” means different things under US federal tax law, Swiss cantonal rules, and German tax law. A retrieval system that does not understand jurisdiction scoping will return a dangerous cocktail of plausibly-relevant but contextually-wrong documents — more harmful than no retrieval at all, because the advisor may not detect the mismatch.
The compliance and audit requirements are non-negotiable. In a professional services firm operating across regulated jurisdictions, every AI-assisted research output must be traceable — not just for quality assurance, but because professional liability means advisors must be able to demonstrate what sources informed their guidance. A RAG system without a structured audit trail is not deployable in this environment, regardless of how accurate its retrieval is.
Sphere Partners built the US Tax Service system with all three constraints as first-class requirements — not as features added after the prototype worked.
About enterprise RAG for professional services
What is enterprise RAG, and how is it different from standard document search?
Enterprise Retrieval-Augmented Generation (RAG) combines semantic vector search with a large language model to retrieve contextually relevant document passages and synthesize a grounded, cited answer — rather than returning a list of document links. Unlike keyword search, RAG understands the meaning of a query and retrieves semantically similar content even when the exact words do not match. In professional services environments, production RAG systems add domain-optimized chunking, jurisdiction-aware metadata filtering, faithfulness evaluation, and audit trail logging that generic implementations omit.
How long does it take to deploy a production-ready enterprise RAG system?
Sphere Partners deployed a production-ready enterprise RAG pipeline for US Tax Services AG in five weeks, covering knowledge source mapping, custom document ingestion pipeline development, embedding configuration, hybrid retrieval architecture, RAG evaluation benchmarking, role-based access control, and audit trail implementation. Timeline varies by knowledge base complexity and number of document sources, but Sphere's Precision-Driven Engineering framework is specifically designed to reduce time-to-production for AI systems in regulated environments.
Can enterprise RAG be deployed securely for compliance-sensitive client data?
Yes — when built correctly from the architecture stage. Sphere's implementation for US Tax Service included role-based retrieval access controls (preventing cross-advisor access to client-specific records), a tamper-evident query and retrieval audit log, and deployment within a data governance framework aligned to the firm's professional confidentiality obligations. Data governance in enterprise RAG is an architecture decision, not a configuration option — it must be designed in from the start.
What retrieval accuracy improvements should firms expect from enterprise RAG?
US Tax Service achieved a 66% improvement in retrieval accuracy compared to their previous keyword-based search approach. Results vary depending on the quality of the existing search baseline, document types, and the specificity of the domain. Sphere benchmarks retrieval precision, recall, faithfulness, and answer relevancy at every implementation milestone — so performance is measured and validated before production deployment, not assumed.
Is enterprise RAG relevant for firms outside the US tax domain?
The architecture Sphere built for US Tax Service applies to any professional services firm managing large volumes of complex, unstructured documents: law firms, audit and accounting practices, financial advisory firms, insurance carriers, healthcare providers, and management consultancies. The core problem — expert knowledge buried in fragmented document repositories, retrievable only by those who already know where to look — is universal to knowledge-intensive professional services.