AI Versus Surveyors: The Right Division of Work

A valuation report can be technically well reasoned and still contain a problem that only becomes obvious when the document is read as a whole. The market value on the executive summary may be £5,000,000 while the conclusion says £5,050,000. A comparable's net internal area may be correct in its own table but wrong when carried into the £ per sq m analysis. These are the practical issues behind the debate about AI versus surveyors.

It is not a contest between professional judgement and software. The useful question is which parts of report review benefit from a systematic second pair of eyes, and which must remain firmly with the registered valuer.

AI versus surveyors is the wrong comparison

A Red Book valuation is not a collection of isolated facts. It is a professional opinion, reached through inspection, analysis of evidence, judgement on assumptions and a clear explanation of how the valuer has arrived at a conclusion. An AI system cannot inspect a property, assess the quality of a letting transaction, decide whether an unusual comparable is truly comparable, or take responsibility for the conclusion.

Nor should it try to. A figure can reconcile perfectly and still be the wrong figure. A market rent may sit comfortably within the range of the evidence but fail to reflect a restrictive lease clause, an inferior specification or an incentive that materially affects the effective rent. Those are valuation questions.

The strength of automated review lies elsewhere. It can read a long draft consistently, compare statements in one section with figures in another, test calculations and check whether stated requirements appear to have been addressed. It does not get tired at the end of a busy day. It does not assume that a number copied from an earlier draft is probably right because it looks familiar.

That makes it a support layer, not a substitute for the valuer. The surveyor remains the author, reviewer and final decision-maker.

Where automated review earns its place

The most useful checks are often not dramatic. They are the small disconnects that arise when a report changes during the course of an instruction: a revised floor area, a late comparable, an amended lease term or a new lender condition.

Figures that do not reconcile

Consider a commercial investment valuation where the adopted rent is amended from £420,000 to £450,000 after further evidence is received. The narrative is updated and the valuation conclusion changes, but the investment calculation still uses the earlier rent. A manual reviewer may spot the discrepancy. Equally, they may focus on the revised comparable evidence and assume the arithmetic has followed through.

An automated audit can identify the two rent figures, find the yield and capital value used in the calculation, and flag that the inputs do not reconcile. It can also identify a market value stated as £3,750,000 in the body of the report and £3,700,000 in the certificate or summary.

The point is not that the system knows the right market value. It does not. It identifies that the report is making two different statements, before the valuer decides which one is intended.

Comparable evidence and unit analysis

Comparable schedules often involve several rounds of adjustment. A property might be analysed at 2,400 sq m in the schedule but described as 2,040 sq m in the narrative. At a capital value of £12 million, the resulting £ per sq m figures can look plausible in isolation while pointing to very different conclusions.

The same applies to dates, unexpired terms, passing rents and yields. If a comparable is said to have 7.2 years unexpired in the commentary but the lease dates indicate 5.2 years, that is worth revisiting. It may be a simple drafting error, or it may reveal that an earlier version of the evidence remains in the report.

A good review tool should test these relationships across the document rather than merely checking whether each individual sentence reads well. That is where report-wide reading matters.

Instruction-specific requirements

Many review issues arise not from valuation methodology but from requirements that must be addressed explicitly. A lender may require a stated marketing period, a minimum level of comparable evidence, or an EWS1 disclosure where it is relevant. A client may have a required form of wording for a particular risk or assumption.

These points can be easy to overlook when a report is otherwise complete, particularly where the instruction changes after inspection. Automated review can check for the relevant disclosure, identify missing sections and flag where the report contains wording that conflicts with the instruction.

It cannot decide whether an EWS1 is required on the facts. That remains for the valuer. But once the instruction requires it, a check that the issue has been addressed is a sensible control.

What must remain with the surveyor

There is a clear line between checking and judging. The valuer determines the appropriate approach, selects and weights comparable evidence, forms a view on covenant strength, interprets lease provisions and decides whether the adopted yield reflects the asset and market at the valuation date.

The valuer also decides whether a flag matters. A report may contain differing floor areas because one is gross internal area and the other is net internal area, with both correctly labelled. A system can draw attention to the difference. The professional must decide whether the explanation is clear and whether the analysis uses the right basis.

This is also why a list of flags should never be treated as a sign-off mechanism. A clear audit result means the system has not found a specified inconsistency or omission. It does not turn a draft into a valuation opinion, and it does not remove the need for a proper human review.

The limits are part of using it properly

AI can misread context, particularly in complex layouts, scanned appendices or reports that use unusual terminology. It may flag an intentional difference. It may not understand the commercial significance of an unusual transaction. A sensible workflow allows the valuer to review, accept, dismiss or investigate each finding.

The quality of the output also depends on the question being asked. A generic document tool may identify spelling or formatting problems, but it will not necessarily understand that a yield, market rent, unexpired term and capital value should tell a coherent valuation story. Sector-specific checks matter because they reflect how valuation reports are actually assembled and reviewed.

Data handling deserves the same care. Reports contain confidential client and borrower information, and firms should understand where files are processed, whether they are stored, and whether they can be used to train models. For valuation work, private encrypted processing with no report storage or model training is not a minor technical detail. It is a basic requirement to establish before introducing any system.

A practical review workflow

The best time to use automated review is when the draft is substantially complete but before it leaves the valuer's desk. At that stage, the system can test the report as one document rather than repeatedly checking unfinished sections.

The valuer should first complete the professional work: inspect, analyse the evidence, settle the approach and write the reasoning. The automated audit then highlights inconsistencies, calculation issues and missing instruction points. The valuer reviews each item in context and makes any necessary amendments. A final read remains valuable, particularly for the conclusion, assumptions and wording that explains the key judgement calls.

This approach is useful for sole practitioners as much as larger teams. In a small practice, there may be no colleague available for a second read on every report. In a larger business, a consistent first-pass audit can help reviewers spend more time on the points that genuinely need professional challenge.

WriteUp was built around this division of work: a valuation-sector audit that checks draft reports in one to two minutes, while leaving every finding and every decision with the surveyor.

The standard worth aiming for

The purpose is not to make reports look machine-checked. It is to give the valuer more confidence that the adopted figure, evidence, calculations and disclosures tell the same story. That is particularly valuable when deadlines are tight and a report has been revised several times.

Professional judgement remains the centre of valuation. Automated review is most useful when it stays in its proper place: catching what a manual review might miss, so the surveyor can focus attention where judgement is required and send out reports that stand up under scrutiny.