VATBuild is not an AI chatbot with VAT knowledge added. It is a three-layer system in which each layer performs a distinct function, and the AI operates only within the constraints established by the other two layers. The distinction matters: the rules are not generated by the AI. The rules are determined first, and the AI works within them.
The rules engine is a proprietary, structured encoding of HMRC Notice 708 and associated notices — VAT431NB and VAT431C for self-builders, Notice 708/6 for energy-saving materials, Notice 701/6 for charities. It operates independently of any AI model, establishing what the correct VAT treatment must be before any AI classification takes place.
At the project level, the rules engine determines the applicable HMRC scheme and the legal basis for any zero or reduced rating — based on building type, previous use, vacancy period, and claimant type. It identifies whether the Domestic Reverse Charge applies, sets the invoicing obligation accordingly, and determines which Notice 708 sections are in scope before a single invoice is processed.
At the line-item level, the rules engine identifies the correct VAT category for each supply, applies the applicable eligibility rules, and constrains the classification to a structured outcome with a mandatory HMRC Notice reference. The AI is required to justify its output against the rules engine's framework — not to generate its own view of what the rules say.
No AI model, however capable, is asked to determine the rules. The rules are determined first. This is the foundation of VATBuild's trustworthiness.
The workflow layer enforces a mandatory sequence of operations that prevents the most common classes of error across both the reclaim and the compliance use cases. This sequencing is not optional — it is enforced at the system level.
The Project Profile must be confirmed before any invoice is processed. The system will not classify line items until the project's VAT position has been formally assessed. This prevents the most common failure mode in DIY VAT management: classifying invoices before the applicable scheme has been correctly identified.
For contractor and developer users, DRC eligibility is assessed at the project level and locked before any supply is invoiced — preventing the most common category of contractor error. Invoice extraction and VAT classification are treated as separate, sequential operations: document reading is completed and verified before any VAT rules are applied.
User review is required before any line item is confirmed. The system distinguishes between AI classification and user-confirmed classification — a critical distinction for HMRC auditability. An independent compliance audit reviews the completed output to catch over-claiming risk before any report is generated. Duplicate prevention is enforced at every stage.
Within this constrained architecture, generative AI performs the tasks it is genuinely well-suited to: document understanding, natural language extraction, and contextualised explanation. VATBuild is model-agnostic — the system is designed to work with any capable AI model and routes different tasks to different models based on capability, cost, and availability.
Invoice document extraction: reading PDFs, photographs, Word documents, and spreadsheets to extract supplier details, line items, amounts, VAT charged, and invoice references. This is a document-understanding task — tolerant of format variation, well-suited to frontier AI models.
Classification justification: generating a plain-English explanation of why a line item has been classified as it has, referencing the specific HMRC Notice section, within the framework set by the rules engine. The AI explains the decision. The rules engine made it.
Compliance audit: an independent AI review of the completed output, specifically to identify over-claiming risk or incorrect rating risk before any report is generated to HMRC or an accountant.
VAT advisor chatbot: answering project-specific VAT questions in plain English, with the project profile and applicable rules in context — so the conversational AI has the grounding it needs to give useful, project-specific answers rather than general guidance.
The AI is not asked to determine which rules apply, to decide what the correct VAT rate is, or to produce output that goes directly to HMRC or a client. Those functions belong to the rules engine and the workflow layer respectively.
Create a free project to see how VATBuild maps your project to the applicable HMRC scheme, enforces the correct workflow, and uses AI where it genuinely helps — not where it creates compliance risk.