RICS has written the AI rules. Now try doing the work.

Consider a reasonably tech-savvy quantity surveyor on an ordinary Tuesday morning. They pay for ChatGPT Pro, have switched off model training and use it alongside Excel and Word. They reconcile a payment application against the previous month, test the assumptions in a forecast, question the meaning of a contract clause and prepare a first draft of a letter. Later, an AI add-in helps turn the cost report into slides for a client meeting.

The files still contain the client’s name, the contractor’s rates and the project references. Removing those details would take time and could break the links needed to understand the records. The surveyor challenges the answers, checks the calculations and decides what to issue. This is not an attempt to hand professional responsibility to a chatbot. It is an attempt to do the work with better assistance.

This is the working day against which an AI standard should be tested. Not just a managed pilot using fictional data, or an unattended system deciding what a contractor should be paid. The immediate question is how an ordinary professional can move between questions, calculations and drafts while using the information the commission requires.

RICS’s October 2026 Surveying skills report makes that question more pressing. It describes a profession short of experienced people, under pressure to improve productivity and increasingly expected to use technology in daily work. Read alongside the AI standard that took effect in March, it raises a question: has RICS explained clearly enough how members are supposed to make that change?

The profession needs the time back

The skills report provides a reason to care about the answer. Some 55% of respondents reported reduced work capacity as an effect of skills shortages, while 50% identified limited time or resources to train existing staff. Meanwhile, 35% selected adopting AI, automation and digital tools as one of their organisations’ strategies for addressing future skills shortages. That last figure covers a group of technologies, not AI alone, but the direction is clear.

The report also recognises that technology is not a substitute for developing people. Its argument is that experienced professionals may be able to oversee more work, while giving more attention to judgement, risk and advice. This is a case for making better use of scarce professional time, rather than assuming that every capacity problem can be solved through recruitment.

There is no contradiction in promoting AI while requiring safeguards. Nor does the report establish that the RICS standard has caused firms to lose time or abandon useful tools. The stronger criticism is about the gap between the ambition and the practical route: RICS is encouraging a change in how services are delivered, while leaving some important conditions of that change open to uncertain interpretation.

That uncertainty has a cost worth examining. A firm cannot confidently build repeat workflows around unresolved questions about what information may be processed, whose authority is needed and how much checking must be recorded. The result may be excessive caution, inconsistent practice or useful work taking place outside the firm’s approved systems.

If shortages are already limiting both delivery and training, the profession cannot afford to treat the time spent meeting these requirements as free. Controls need to protect clients, but they should also be tested for whether they preserve the benefits of the work they regulate.

The proposal becomes the problem

I recently tried to put a practical AI arrangement into a consultancy proposal. The firm expected to use approved systems, protect project information, check the results and remain responsible for its advice. The client would know this in advance and could exclude particular uses. The wording needed to cover pre-contract work, post-contract services, claims and cost consultancy without a fresh negotiation for every task.

RICS requires advance disclosure through the documents governing the appointment. This includes intended uses, available professional indemnity cover, complaints and redress, and any option to opt out. A proposal is therefore a sensible place to establish the arrangement. There is no need to turn every later use of the agreed system into another announcement to the client.

The difficulty appeared when the drafting reached project information. Section 3.1 requires “express written consent in advance from affected stakeholders” before private and confidential uploads, alongside an assessment of the risk. The standard does not define that group. RICS’s client information note describes obtaining the client’s consent, leaving the consultant to work out how the two positions apply when the client supplies records from several parties.

A disclosure clause, a consent clause and a warranty about third-party permissions do different jobs. The first explains the proposed use. The second grants permission within the client’s authority. The third allocates responsibility for permissions that may be needed elsewhere. Combining them into a longer clause does not necessarily settle whether the authority exists.

This matters because a standard proposal needs to establish a repeatable commercial arrangement. It cannot become a promise to investigate the permission history of every document the client might supply. Equally, the client should not be asked to sign an assurance whose meaning neither party can confidently explain.

The standard does not expressly demand a signature for each invoice. The problem is that its wording leaves firms to decide how far short of that exercise they may reasonably stop.

A claim is not a collection of consent forms

Consider a claim containing 50,000 records. There are invoices, programmes, subcontract accounts, labour records, quotations and years of correspondence. Some documents came from the contractor, some from suppliers and some from the employer whose conduct is being challenged. The consultant needs to examine them as a connected body of evidence, not as thousands of unrelated files.

A supplier name may be needed to identify duplicate invoices. A reference may connect a cost to a work order. An instruction may explain why the forecast changed. Removing identifying details can sometimes protect information without affecting the analysis. In other cases, it removes the relationships that make the analysis possible.

The distinction should be between information the task needs and information it does not. A complete payroll may be unnecessary for a package-level forecast. Invoice references may be essential to a reconciliation. Treating both as a general instruction to remove client-specific information is neither a useful analytical method nor a complete privacy assessment.

The phrase “affected stakeholders” leaves another distinction unresolved. Is a party affected because its privacy or confidentiality interests are engaged by the processing, because it created a document, or because the resulting assessment may reduce its claim? A commercial interest in the outcome is not the same question as a right over the handling of the evidence.

Historic claims make this more awkward. A subcontractor may have been dissolved, the relevant staff may have left and nobody may be available to provide fresh permission. The records still need to be assessed. The sensible starting point is to examine the rights and permissions that already apply, not assume that a new signature is required because the analytical tool has changed.

RICS’s wider data-handling guidance already recognises contractual rights, licences and consent where needed. Its AI guidance should explain how those principles apply to complete submissions, records from several parties and confidential processing services. Firms need to know when they may rely on the client’s authority, what evidence they should retain and which circumstances require further enquiries.

I would want a clear process for assessing authority at the level of the submission or class of records. The client would explain its rights and disclose restrictions. The consultant would make reasonable checks and investigate specific concerns. Existing written permissions would be considered on what they actually allow.

That would not require pretending that the client owns every document. It would distinguish a proper enquiry from an exhaustive reconstruction of the project’s history. A longer proposal cannot supply that distinction if the professional requirement itself remains unclear.

Training off is a start, not a verdict

The surveyor’s training setting is a real safeguard. OpenAI states that switching off model improvement prevents new conversations from being used for training. It also explains that this does not delete saved chats, and that voluntarily providing feedback can allow the associated conversation to be used despite the general opt-out.

Training is therefore one question, not the whole assessment. Who controls the account? What information can connected applications receive? How are records retained? What happens when an employee leaves? What contractual restrictions apply to the project? Those matters should be settled through a firm’s approved arrangements, rather than left to each surveyor to discover while finishing a report.

Subscription labels are an incomplete answer. A more expensive individual account does not, by itself, establish stronger contractual protection. Equally, suitable business arrangements are not necessarily confined to bespoke enterprise purchases. OpenAI offers data-processing agreements for ChatGPT Business and its API, as well as Enterprise.

The firm should assess the actual service, settings and intended use. Where an individual paid account lacks a necessary protection, management should provide a suitable route to the same work. The answer should not be to leave staff choosing between a useful tool and a policy that offers no practical alternative.

James Garner’s contribution to the skills report is relevant here. He warns against long commitments to products that may add little to general-purpose models organisations already pay for. He also argues that AI literacy involves judging the risks and understanding professional requirements, including the RICS standard. His position is not that ordinary subscriptions make governance unnecessary. It is that buying another product is not the same as understanding what the practice needs.

That distinction should shape the guidance. Firms should be able to explain which protection a tool adds, which risk it addresses and why it is necessary for the work. A specialised surveying label is not evidence of suitability any more than a familiar provider name is.

Approve the work, not every prompt

RICS has already made some allowance for scale. Its standard focuses on material effects on services, permits standing firm-level assessments and allows sampling for automated or high-volume outputs. It would be unfair to suggest that every prompt requires a separate approval.

The less obvious case is the ordinary working conversation. A surveyor asks about a clause, challenges the answer, supplies an amendment, rejects a suggested argument and develops a different draft. Which part is the output relied upon, and what record should show that the eventual professional decision was sound?

I would organise the record around the piece of work and the advice adopted. For a reconciliation, retain the source versions, calculations, unresolved differences and review decision. For a forecast, retain the assumptions and the reason for selecting the final position. For a contractual letter, retain the verified terms, relevant facts and approved draft.

The working papers should address the required reliability decisions without creating a second administrative account of the same exercise. A saved conversation can help explain what happened. Its existence does not establish that the work was checked, and its length is not a measure of professional care.

Checking must also reach beyond whether the final answer looks plausible. A sample of extracted invoice lines will not establish that every intended attachment was included. A forecast can calculate correctly while allowing for the wrong staffing period. A summary can accurately describe the documents selected while omitting the instruction that changes the assessment.

None of this requires repeating the whole task by hand. File counts, reconciled totals, targeted tests, suitable samples and review of exceptions can provide evidence that the process worked. For repeated tasks, useful tests should become part of the workflow. The objective is a result that can be defended, not proof that somebody could have reached it more slowly.

The controls should also reflect the task’s consequences. Extracting dates into a checked reconciliation and issuing a payment decision may both materially affect the service. They do not involve the same authority or exposure. Material contribution brings a use within the standard; it should not replace a proper assessment of what could go wrong.

Agents need defined authority

The skills report goes beyond ordinary chat use. Tom Heath’s contribution describes a future in which experienced surveyors supervise AI agents handling repetitive work, leaving more time for valuable advice and developing junior colleagues. These are his stated views within the report, not a new permission issued by RICS, but they describe the working model the profession is being encouraged to consider.

Apply that model to a payment review. A submission arrives, attachments are saved, schedules are compared with the previous application and draft queries are prepared. A further step updates a working register. Another might send the queries. The process may cross several tools without the surveyor manually uploading a file at each stage.

The important decisions concern what information each step may reach and which actions it may take. There is a considerable difference between preparing an assessment and issuing it, or between reading a ledger and altering it. Approval of the final report may be too late if an earlier step has already disclosed information or changed a live record.

The same principle applies to add-ins within Word, Excel and PowerPoint. The familiar window does not tell the firm which service handles the information or what authority the tool has. Approval should follow the actual processing arrangement, including connected services and access to other project files.

I would require a workflow to distinguish evidence from instructions, preserve original records and obtain approval before actions with contractual or financial consequences. A contractor’s submission should be material for the consultant’s system to assess, not a source of authority over how that system behaves.

This is where stronger controls earn their keep. Reading, drafting, updating internal records and communicating externally need different boundaries. Repeated permission requests for ordinary steps are not a substitute for deciding where those boundaries belong.

If agents are to relieve pressure on scarce professionals, the operating rules need to work at the level of the authorised process. Otherwise, the automation may move the information faster while leaving the surveyor to resolve the same unanswered permission question at every stage.

Make the permitted route usable

For the surveyor at the start of this article, I would build an approved daily workflow rather than an exception procedure that occasionally permits AI. The account and tools would be assessed at firm level. The appointment would describe the purposes, handling arrangements, relevant authority and exclusions. Staff would know the permitted tasks and the circumstances requiring further review.

Within that arrangement, ordinary use should proceed without reopening the whole assessment. A change from reading a register to editing it deserves attention. Another question about the same contract, using the same authorised records and safeguards, should not require a new negotiation. The trigger should be a meaningful change in purpose, information, access or authority.

Client communication should work on the same basis. Explain the expected use across the commission and obtain the appropriate written agreement. Record specific restrictions and establish how objections will be handled. Provide a clear account of professional responsibility and available insurance. Then distinguish those appointment-wide matters from the checks required on individual pieces of work.

RICS could make this easier through model wording that addresses mixed-party records directly. Firms need more than a broad AI disclosure clause. They need worked examples showing when client authority is sufficient, how existing permissions should be assessed and what reasonable enquiries look like for historic claims.

The institution could also reduce repeated supplier work. Section 4.1 requires written requests for information including environmental effects, training data and supplier liability, with missing information treated as a recorded risk. A maintained set of supplier information, and clearer guidance on reliance on published material, could prevent small firms from repeating the same enquiries without improving the answer.

That would not remove the firm’s responsibility to assess its own use. It would allow the firm to spend more time testing the workflow and less time collecting information another practice has already requested. The skills report’s findings on limited time and resources make this an obvious place to examine the burden.

Training should follow the work too. Tim Shelton’s contribution to the report argues that, as traditional graduate tasks are increasingly automated, training needs to be redesigned rather than graduate intake reduced. Earlier exposure to judgement, client discussions and data interpretation is a useful direction. It also requires approved access to real work, with suitable supervision, rather than leaving junior staff permanently confined to fictional examples.

A cost consultancy should be able to teach reconciliation, forecasting and contractual analysis through controlled AI-assisted exercises. The assessment would ask whether the surveyor understands the records, can explain the calculations and can identify where the tool failed. Professional knowledge remains essential; it is what makes useful review possible.

The firm also needs a practical response to existing unapproved use. Staff should be able to disclose the tools they have used, establish whether information needs protection and move the work into an approved arrangement. A policy that encourages concealment would leave management less able to protect the client, however strict it appeared on paper.

Test the rules against an actual commission

A standard can be sound in principle without being sufficiently clear in application. The answer is not to remove professional duties because they take time. It is to test whether the way those duties are expressed directs effort towards the risks that matter.

RICS should put the guidance through a small cost consultancy with no separate technology department. Give it a normal paid subscription, a mixed-party claim, a monthly reporting cycle and an agent with limited access to project records. Ask it to demonstrate authority, safeguards, checks and client communication. Then measure the work needed to do so and examine where competent practitioners reach different answers.

The test should include errors avoided and time saved, as well as the records produced. If the approved workflow catches duplicate costs, exposes forecast assumptions and leaves a better audit trail, those benefits belong in the assessment. The manual alternative should be examined, not treated as an error-free benchmark simply because it is familiar.

Some improvements are available now through standing approvals, better appointments and proper working records. Others require clearer guidance or changes to the standard. A case for improving the rules is not a claim that every current workflow already complies with them.

My view is that the RICS standard remains a useful statement of responsibility, but not yet a sufficiently clear and scalable framework for everyday AI-assisted surveying. The skills report strengthens the case for closing that gap. It explains why the profession needs more capable use of technology, while the unresolved questions in the standard affect whether firms can make that use routine.

The line should be drawn around unauthorised disclosure, unsupported advice and uncontrolled action. It should not be drawn around whether a competent surveyor has found a faster way to prepare the work. RICS’s skills report makes the case for change. Its professional guidance now needs to make the responsible route clear enough to follow on an ordinary Tuesday morning.