Generic AI assistants are genuinely useful for explaining how construction VAT works in general terms. They are structurally insufficient for making real compliance decisions on real invoices. This comparison shows the gap — across the dimensions that actually matter when an HMRC enquiry arrives.
The stakes in construction VAT are concrete. A self-builder with a £200,000 conversion project typically has between £15,000 and £40,000 of reclaimable VAT. A contractor issuing invoices at the wrong rate across a year's work faces HMRC assessment going back four years. A developer miscategorising mixed-use supplies creates liability on both sides of every affected transaction.
In each case, the question is not whether AI is useful — it is. The question is whether AI alone is sufficient to produce a trustworthy, defensible, auditable compliance decision. The comparison below shows why it is not.
The 14 dimensions below are drawn directly from the technical architecture brief. Each one represents a real difference in outcome — not a marketing distinction.
Statistical probability — the most likely response given training data
Deterministic rules engine — HMRC Notice 708 encoded explicitly; outcomes grounded in law, not probability
Cannot distinguish between the reclaim obligation (VAT431) and the ongoing invoicing obligation (Notice 708)
Handles both — different workflow and output for each user type: self-builders, developers, contractors, suppliers
None — the model has no knowledge of your project unless you re-state every relevant fact each turn
Full project profile conditions every classification decision: building type, prior use, vacancy period, DRC status
May mention DRC in general terms; cannot determine whether it applies to a specific supply or transaction
DRC eligibility assessed and locked at project level before any invoice is processed
Training cutoff — may not reflect recent HMRC updates, clarifications, or rate changes
Rules curated and maintained against current HMRC notices — not derived from training data
Citations generated probabilistically — section numbers and notice references may be hallucinated
HMRC section references validated against the rules engine — not generated freely by the AI
Not possible — the model operates on prose and cannot process structured invoice data
Every line item extracted, categorised, and individually classified against project-specific rules
Unstructured prose — cannot be ingested downstream or attached to an HMRC claim as evidence
Structured classification exportable to CSV, Excel, and PDF; matched to HMRC report fields
None — the conversation is ephemeral; there is no retained record of any decision
Every classification, override, and decision timestamped and stored — the full record available at any time
Not possible — prose output cannot be converted into a valid VAT431 or contractor compliance report
Reports match VAT431NB/431C fields; contractor reports reference Notice 708 sections throughout
None — the model has no mechanism to catch its own errors or assess the risk of its recommendations
Independent compliance audit reviews every output before it is released; flags over-claiming risk
None — the model has no knowledge of what has been claimed previously
System tracks which line items have been included in previous reports and blocks re-inclusion
Non-deterministic — the same input may produce a different output on re-run; no stable baseline to test against
Consistent and reproducible — the same invoice against the same project produces the same classification, every time
No party takes responsibility for the output; no mechanism to challenge or verify it
Every decision traceable to a specific HMRC rule; user confirms each one; full record available under enquiry
Create a free project and upload your first invoice. See how VATBuild's rules engine, workflow, and targeted AI produce a classified, HMRC-referenced output — not a probabilistic answer.