The Royal Institution of Chartered Surveyors’ (RICS) 2026 survey of more than 3,100 chartered surveyors found that two-thirds of construction professionals and three-quarters of those in commercial property now use AI, but most of that use is still at the pilot stage. Since 9 March 2026, the RICS professional standard on the responsible use of AI in surveying practice has required members and regulated firms to apply and document professional judgement over AI outputs that materially affect their work. The standard sets out what firms must do. It says less about how to do it day to day.

On 19 October 2026, the RICS published How to build a judgement layer for AI use in surveying, an insight paper written to fill that gap. Chris de Gruben FRICS, Senior Director at Artefact UK and co-chair of the working group behind the standard, co-authored it with Anil Sawhney FRICS (RICS), James Garner FRICS (Gleeds) and David Wong MRICS (The Augmented Practitioner Column).

Why agents change where judgement is applied

With a chatbot, the surveyor frames the question at the start and judges the answer at the end. An AI agent is given a goal, tools and permission to act, and chooses its own steps. Errors can enter long before the output arrives: the wrong evidence retrieved, the wrong tool called, an assumption carried forward unchecked. The paper gives the example of an agent assembling comparables that pulls in two from an adjoining submarket with different demand and yields. The draft valuation looks plausible and is wrong, and a review of the final output will not show why.

Judgement therefore applies to the process as well as the result. Deciding what an agent may do, what evidence it may use and where it must hand control back to the surveyor is itself a professional decision.

Connect, codify, govern

The paper’s answer is a judgement layer: the way a firm organises its information, knowledge and controls so that AI-assisted work can be reviewed and signed off properly. Firms build it through how they organise and record their work, rather than buying it as software. It has three elements.

Connect. The AI system needs access to the evidence a surveyor would normally use, and the surveyor needs to see what it relied on. Firms do not need perfectly clean data to start, but they do need to know where their main evidence sits and what the system can reach.

Codify. This records how that evidence should be treated: which comparables are admissible, when an assumption needs a second review, when a job goes to a senior colleague. A firm’s knowledge base grows case by case, so this is the slowest element to build.

Govern. This turns the firm’s rules into controls: what the system may do, where it stops, what is checked and who signs off. It produces two records, a trail of what the system did and a record of the named surveyor’s decision to rely on, change or reject the output.

In many firms these controls sit partly inside vendor tools, such as suggested comparables or AI-drafted sections of a survey report. The vendor has not seen the client’s instruction and does not sign the report, so the paper treats vendor defaults as design choices for the firm to specify and test.

Artefact’s contribution

Artefact brought the perspective of a firm that deploys AI in property businesses day to day: what works and what does not, how agents are rolled out in practice, the challenges teams run into along the way and the impact on how the work gets done. That experience grounds the paper in what firms can do now, with the systems and data they already have.

“Most of the AI projects we see in property do not fail because the model is weak. They fail because the firm has not decided where the evidence lives, what the rules are and where a human has to step in. The judgement layer gives surveyors a common language for those decisions, and it is the same discipline we apply when we build agents for clients.” – Chris de Gruben FRICS, Senior Director, Head of Property UK & EU, Artefact

Where to start

No firm needs the whole layer in place before using AI. The paper recommends starting with one recurring workflow where the output could influence advice:

  • identify the evidence normally used and check what the AI system can access;
  • write down one or two recurring decisions and define where the system must stop for review;
  • test the workflow on real examples and adjust the rule, the data or the control when exceptions arise.

For a sole practitioner, connect might mean moving condition records into a folder the AI system can read, and govern might be a short note of what was checked before the output was used. The RICS standard’s requirements, including the AI systems record and supplier due diligence, still apply at any size.

The paper also addresses training. Routine work has long been how junior surveyors learn, by assembling evidence, spotting exceptions and watching senior colleagues challenge a first answer. If AI absorbs that work and nothing replaces it, firms gain efficiency now and lose capability later. The paper recommends supervision that lets juniors frame the task, inspect sources and question assumptions before a senior surveyor signs off.

Read the full paper on the RICS website.

About the author

Chris de Gruben FRICS is a Senior Director at Artefact, where he leads the UK property team. He co-chaired the RICS working group that wrote the Responsible use of AI in surveying practice professional standard, is Vice Chair of the RICS Valuation Professional Group Panel and is co-authoring the forthcoming RICS guidance on AI in real estate valuation.