Building code compliance checking is one of those responsibilities that absorbs far more time than it should. Recent industry data suggests that compliance-related work costs UK practices roughly £16,700 per employee each year. That figure covers not just the checking itself but the rework, the coordination gaps, and the hours spent chasing product information that should have been accessible from the start. AI-assisted compliance tools are beginning to offer a different approach. The question is whether the technology is mature enough to trust with something this consequential.
If you haven't sat through one recently, it's worth walking through what a typical compliance review looks like in practice. Take a mid-size residential project at Stage 4. The architect or technologist pulls up the Approved Documents relevant to the building type. For a new-build dwelling, that's a minimum of Parts A, B, C, F, L, M, O, R, and S. Each document references specific performance criteria, acceptable construction details, and standards like BS 9991 or BS EN 13501 for fire, or BR 443 and SAP calculations for energy.
The technologist then works through the design drawings, specification clauses, and material schedules, checking each element against the applicable requirements. Does the external wall assembly hit the target U-value? Do the fire compartmentation details match the strategy in the fire safety report? Are the accessible thresholds detailed correctly? Is there a conflict between the acoustic separation required by Part E and the ventilation strategy in Part F?
Each of these checks requires the reviewer to hold multiple documents in mind simultaneously, cross-referencing between the specification, the drawings, the product data sheets, and the regulatory text. A single inconsistency between a window schedule and the Part L calculations can trigger a Building Control query that delays approval by weeks. Multiply that across every building element, and you start to see why compliance absorbs so much of a project's fee.
The problem with manual compliance checking isn't competence. Most architectural technologists know the regulations well. The problem is volume. A typical Stage 4 package for a medium residential scheme might include 80 to 120 drawings, a specification running to several hundred pages, and dozens of product data sheets. No human reviewer can hold all of that in working memory while simultaneously parsing regulatory text.
The failures tend to cluster in three areas. First, coordination gaps between documents. The specification says one thing about a cladding system's fire rating, but the drawing detail shows a different build-up. Neither is wrong on its own, but together they create an inconsistency that Building Control will flag. Second, version drift. The energy calculations were based on Rev C of the window schedule, but the specification was updated to Rev D with a different glazing unit. The U-values no longer align, but nobody noticed because the documents were edited in different software at different times. Third, interpretation differences. Approved Documents are guidance, not law. Two experienced technologists can read the same clause and reach different conclusions about what constitutes compliance, particularly in areas like means of escape where professional judgement plays a large role.
These aren't rare edge cases. A 2023 RIBA survey found that regulatory coordination issues were among the top three causes of project delays reported by practices. The Building Safety Act has only increased the pressure, adding new requirements for higher-risk buildings that demand more documentation, more sign-off, and more rigorous record-keeping throughout the design process.
AI compliance tools work by ingesting both the regulatory text and the project documentation, then systematically checking for conflicts, gaps, and non-conformances. The better systems don't just pattern-match keywords. They parse the semantic structure of regulations and map them against the structured data in specifications and schedules.
Consider how this works for a Part L check. The AI reads the specification's thermal performance data for each building element, pulls the declared U-values from product data sheets, and compares them against the notional building values in Approved Document L. If the specified glazing unit has a centre-pane U-value of 1.2 W/m²K but the energy model assumes 1.0 W/m²K, the system flags the discrepancy before it reaches Building Control. That kind of cross-check is exactly what a human reviewer would do, but the AI does it across every element simultaneously, in minutes rather than days.
Platforms like Avoice take this further by connecting compliance checking to the specification writing process itself. Because Avoice generates specifications classified under Uniclass and CAWS standards, and ingests a firm's existing project documentation, material libraries, and historical data, it can flag inconsistencies between the spec and other project documents before they become problems on site. The compliance check isn't a separate activity bolted on at the end. It's embedded in the workflow from the start.
The numbers are striking. Tools like PlanCheckPro report that reviews which typically take building departments weeks can be completed in the same day using AI. Even accounting for the human review that should follow any automated check, practices using AI-assisted tools report cutting their compliance review time by 60 to 70 percent.
For a small practice running three or four projects simultaneously, that time saving translates directly to fee recovery. If compliance-related work absorbs £16,700 per employee annually, and AI tools can reduce that by even half, the return on a software subscription is measured in weeks, not years. More importantly, the reduction in Building Control queries and rework means fewer delays, fewer awkward conversations with clients, and fewer of those painful moments where you realise a regulatory conflict should have been caught two stages ago.
There's a second-order benefit too. When compliance checking is faster and more reliable, it changes how practices approach early design stages. Instead of deferring compliance considerations until Stage 4, teams can run preliminary checks during Stage 2 or 3, catching issues while the design is still flexible enough to accommodate them without significant rework. Avoice's approach of generating specs that cite the right standards and clauses, grounded in the firm's own data rather than generic clause libraries, supports this kind of early-stage validation.
It would be irresponsible to suggest that AI can replace professional judgement in compliance checking. It can't. The Approved Documents are guidance documents that require interpretation, and interpretation requires experience, context, and an understanding of what Building Control officers in a particular local authority tend to accept or challenge.
Means of escape design is a good example. Part B gives prescriptive guidance for common building types, but anything unusual, a mixed-use scheme with a shared basement car park, for instance, often requires a fire-engineered approach. The AI can verify that the specified fire door has the right rating and that the travel distances fall within the prescribed limits. What it can't do is assess whether the overall fire strategy makes sense for the specific building configuration, or anticipate the questions a Building Control officer might raise about a non-standard layout.
Similarly, AI tools currently work best with structured data: specifications, schedules, tabulated performance values. They're less effective at interpreting design intent from drawings, particularly complex details where compliance depends on how components interface with each other in three dimensions. A cavity barrier detail that looks compliant in a 2D section might fail in practice because of how it meets the window reveal. That kind of spatial reasoning still requires a trained eye.
The most productive approach treats AI as a first pass. Let it handle the volume, the cross-referencing, the systematic comparison of values across documents. Then apply professional judgement to the areas where context, experience, and spatial understanding matter most.
The market for AI compliance tools is still young, and the options vary significantly in scope and approach. Some tools focus narrowly on specific regulations or building types. Others try to cover the full regulatory landscape but sacrifice depth for breadth.
For UK practices, the most relevant consideration is whether the tool understands the structure of the Approved Documents and the classification systems used in British specifications. A tool trained primarily on the International Building Code won't parse Part B the way a UK-trained system would. Avoice is built specifically for architectural workflows and generates output structured around recognised UK standards including Uniclass and CAWS, which means its compliance flagging is grounded in the same classification framework that Building Control expects to see.
Integration matters too. A compliance checker that requires you to export, reformat, and upload your project data is adding friction to a process that's supposed to reduce it. The tools that deliver the most value are those embedded in the specification and documentation workflow, catching issues as the documents are written rather than after they're finished.
Several jurisdictions are already moving toward formal recognition of AI-assisted compliance checking. In the US, Florida's House Bill 683 authorises private providers to use automated plan review systems for code compliance, and Texas and Tennessee have expanded similar provisions. In Europe, draft addenda to Eurocode are expected to pilot in select jurisdictions this year. The UK hasn't formalised equivalent provisions yet, but the direction is clear.
For architectural technologists, the practical implication is straightforward. The practices that adopt AI-assisted compliance checking now will spend less time on regulatory administration and more time on the technical design work that actually requires their expertise. The practices that wait will continue losing days to manual cross-referencing, absorbing the cost of rework when inconsistencies slip through, and watching their more efficient competitors win work on tighter fees.
The regulations aren't getting simpler. The Building Safety Act has seen to that. And the documentation burden will only increase as performance standards tighten and accountability requirements expand. Building your compliance workflow around manual review alone isn't just slow. It's a risk that gets more expensive with every regulatory update. If you want to see how AI-assisted specification and compliance checking works on a real project, Avoice is worth a look.
