Guides
How AI actually gets implemented in UK construction
Reference guides rather than commentary. Each one answers a single question, cites what it stands on, states where the approach fails, and carries the date it was last read against its sources.
Where a subject turns on law, contract, data protection or site safety, the guide says so and names the specialist who reviewed it. Nothing here is advice on a specific project.
Flagship course · 14 of 14 parts published
The Construction AI Syllabus
Fourteen parts, one a day, for UK construction professionals. Every published part on one page, read in order or jump straight to the part you need, switched by tab with nothing to reload.
Start the course
Value, time and return
The Construction Automation Audit
A method for walking your own business to find the automation worth doing: six questions per task, a judgement test, and the question most audits never ask.
8 sources
How to Calculate the ROI of Construction AI
Build a return you can defend: observed hours, a fully loaded cost from UK statutory rates, review time, and three adoption cases. With a worked example.
10 sources
Construction Processes You Can Automate Today, Ranked by Evidence
Ranked by what vendors document rather than advertise: rule-bound processes automate cleanly, extraction needs review, and drawings remain unevidenced.
13 sources
Where AI Saves Time in UK Construction, and Where It Does Not
AI saves time turning site messages, photographs and paperwork into records. It does not read drawings. What the ONS data shows, and what no UK source shows.
12 sources
Getting started safely
How to Run a Safe 30-Day Construction AI Pilot
Five things fixed before day one, then the four weeks. What actually binds a UK contractor is UK GDPR, and the RICS AI standard if your QS team is regulated.
15 sources
Using ChatGPT, Gemini and Claude in a UK Construction Business: A Practical Guide
The free tier open in most people's browser tabs is the only one of nine account types across the three vendors with no data processing agreement available at any price. RICS now has a mandatory standard on this. Here is what actually changes by tier, what the evidence says these tools are and are not reliable at, and what to check before you paste in a site photo.
20 sources
Tools and comparison
Microsoft 365 Copilot or a Custom Construction AI Assistant?
Copilot costs £23.10 per user monthly on top of Microsoft 365. What it reaches, the shared mailbox and file format limits, and when a build is justified.
11 sources
Why Am I Paying So Much for Software I Don't Use?
The usual answer is that you have shelfware. The stronger, better evidenced answer is that the contract decides the bill, not the user: seats ratchet up but not down, and the walk-away right a government buyer gets on 30 days' notice is exactly the term you were never offered.
13 sources
Can I Tailor-Make Procore, Asite or 4Projects to My Business?
Up to a point, and the point is narrower than the sales conversation suggests. What you are actually choosing between is which fixed data model to accept, because the configuration surface on every major platform is smaller than it looks.
12 sources
Your Business, Your Construction Platform: One Product or Something Bespoke?
Every platform that markets itself as one system quietly advertises 400 or 500 integrations for what it does not do. The honest decision is not buy versus build. It is which of your processes are genuinely yours to protect, and which just came from the last system you happened to use.
11 sources
How to Get Microsoft to Talk to Your Construction Platforms
There is no Microsoft-built connector for Procore, Autodesk or Aconex. What you actually get is a mix of third-party bridges of very different provenance, and Microsoft's own documentation is the most honest place to learn what each one is really built on.
12 sources
Governance, security and risk
Foundations
How the systems are actually built
AI and Software Architecture Translation Guide
The full six part guide to how modern AI systems are assembled, for construction leaders and teams deciding on them without becoming software engineers.
5 sources
Where AI Belongs, and Where Code Belongs
The one decision that sets an AI project's cost and reliability: when a correct answer can be written as a rule, and when it needs real judgement.
3 sources
The Model Landscape, Safeguards and Access in 2026
A snapshot of AI model tiers, pricing and access controls as at 17 August 2026, valid to 30 November 2026. Capability and account access have split apart.
6 sources
Under the Bonnet: Waiting, and Where Data Lives
Why an AI system stands still or keeps working while it waits, and how PostgreSQL's JSONB lets a controlled register hold data that varies by trade.
3 sources
How a System Finds the Right Record and Cites It
How a system finds the right project record and cites it: metadata, embeddings and vector search explained, with the limits nobody usually quotes.
3 sources
Agents, Tools and Orchestration
What an AI agent is actually allowed to do, who decides, and the eight layer stack turning a chat window into a controlled business process.
4 sources
Putting It to Work: Routing, a Worked Example, Where to Start
A model routing matrix for construction work, a nine stage worked example from voice note to portfolio answer, and where to start building it.
3 sources