Rules Based Versus AI Auditing for Valuers

A report can be technically sound, well reasoned and still contain a figure that does not reconcile. A market value of £2,500,000 may appear in the executive summary while £2,550,000 remains in the valuation section. The difference may be a late amendment, not a valuation error. But once it reaches the lender, the cause matters less than the contradiction.

That is the practical question behind rules based versus AI auditing. Both approaches can help a valuer review a draft before issue. They do different jobs, however, and treating one as a substitute for the other can leave gaps.

What rules-based auditing does well

Rules-based auditing works from defined checks. If a lender instruction requires a marketing period, the system can confirm whether the relevant wording appears. If a rent is entered as £125,000 per annum in one field and £115,000 elsewhere, a numerical comparison can flag the difference. If a floor area is stated in square metres but not in square feet where both are required, that too can be checked against a set rule.

This has real value. The checks are predictable, repeatable and easy to explain. A firm can set a requirement once and apply it consistently across a team. For recurring lender or client criteria, that consistency is useful, particularly where a reviewer needs to establish whether a required item is present rather than debate its quality.

Rules work best when the expected answer is clear. Is the inspection date present? Does the report name the client? Has the stated unexpired term been entered? Does a calculation produce the stated figure when the inputs are known? These are precise questions.

The limitation is that valuation reports are not structured forms. They are professional narratives containing assumptions, caveats, evidence and judgement. The same point may be expressed in several places, using different language, and may only become significant when read alongside another section.

Where rules-based checks can fall short

A fixed rule can identify a missing phrase. It is less able to assess whether the surrounding explanation is internally consistent.

Take a retail investment valuation. The report may state that the property is let on a ten-year lease with five years unexpired. Elsewhere, the tenancy schedule records a lease expiry date that leaves four years and two months. A rules-based check may identify the dates if it has been configured for them. It may not recognise that the narrative conclusion about risk, covenant strength and adopted yield relies on the longer term.

Comparable evidence creates a similar problem. A rule can count the number of comparables cited. It cannot necessarily judge whether the report says Comparable 2 supports a rate of £320 per sq m while the table records £230 per sq m, or whether an adjustment described in the narrative has not been reflected in the analysis.

This is not a criticism of rules. It is a question of fit. Rules are good at confirming defined requirements. They are less suited to the contradictions that arise when a report changes during drafting, when a schedule is updated without the commentary following it, or when a conclusion is supported by evidence that has been described inconsistently.

Rules based versus AI auditing: reading the report as one document

An AI-supported audit adds a different layer. Rather than only asking whether a prescribed field or phrase exists, it can compare information across the report and flag points that appear not to align.

For example, it may identify that a report describes a passing rent of £78,500 per annum in the tenancy section, but uses £87,500 in the investment calculation. It can flag a yield of 6.25% in the valuation rationale and 6.50% in the calculation appendix. It can also identify a stated net internal area of 1,250 sq m where one comparable analysis uses 1,520 sq m without explanation.

The important word is flag. The system does not decide which figure is correct. There may be a valid reason for each apparent discrepancy: a different area basis, a rent-free adjustment, a rounding convention or a late instruction change. The valuer reviews the finding, checks the source material and decides whether the report needs amendment.

That distinction matters. AI is not there to form the opinion of value, choose comparable evidence or replace the valuer's judgement on market conditions. It handles repetitive cross-checking at speed, giving the surveyor a clearer view of points that merit attention before the report leaves their desk.

AI can also help with report-wide requirements that are difficult to reduce to a single keyword. An EWS1 disclosure, for example, may be present but conflict with the property description, the valuer's assumptions or the lending basis. A marketing-period statement may appear in the report but not address the particular requirement set out in the instruction. These are the areas where reading context matters.

The strongest process uses both methods

For most valuation teams, this is not an either-or decision. A sensible review process combines deterministic checks with contextual review.

Start with the non-negotiables. Client-specific requirements, lender thresholds, mandatory disclosures and standard report fields should be checked consistently. Rules are well suited to this work because the criteria are known in advance. They provide a clear record that the required points have been considered.

Then use AI-supported review to search for inconsistencies across values, rents, areas, yields, dates, lease terms and comparable evidence. This is particularly helpful after late changes. A revised market rent can affect the valuation calculation, the executive summary, the comparable discussion and the sensitivity analysis. It is easy for one reference to remain unchanged, even in a carefully prepared report.

Finally, the valuer performs the professional review. That is where the report is tested against the instruction, inspection findings, market evidence and the valuer's own reasoning. Neither a rule nor an AI finding can determine whether an assumption is appropriate. Both can make the final review more focused.

A practical example

Consider a draft valuation of a multi-let industrial estate. The capital value is amended from £6,750,000 to £6,700,000 after a review of void allowances. The summary, valuation certificate and calculation may all update correctly. The narrative may still say that the adopted equivalent yield is 5.75%, while the revised cash flow produces 5.90%.

A rules-based check can confirm that the certificate contains a market value and valuation date. An AI audit is more likely to flag that the stated yield and the revised analysis do not appear to reconcile. The reviewer can then decide whether the narrative, calculation or both need correcting.

The same principle applies to minimum comparable evidence. A rule can establish that three comparables have been included where three are required. An AI-supported review can help identify that one of those comparables is described as freehold in the table but as leasehold in the commentary, which may affect the weight given to it.

When rules may be enough

A short, highly standardised report with limited narrative may benefit mainly from defined checks. If the instruction is simple, the data is held in structured fields and the principal risk is omission, a rules engine can provide an efficient first pass.

Rules are also useful where a firm wants absolute consistency around its internal presentation and required wording. They do not need to interpret nuance to confirm that a specified statement has been included.

The case for AI-supported auditing becomes stronger as reports become longer, more bespoke or more heavily amended. It is also stronger where a review team needs to assess volume without reducing the time given to genuinely professional questions. The more places a key fact appears, the greater the value of checking whether those references still agree.

Confidentiality and control still come first

The audit method should not create a new concern around confidential client information. Before using any system, firms should establish where report data is processed, whether it is encrypted, whether it is retained and whether it is used to train external models.

A valuation report often contains commercially sensitive details. That calls for private, encrypted processing, no unnecessary data storage and no use of submitted reports for model training. It also calls for clear user control: findings should be visible to the surveyor, reviewable in context and capable of being accepted or dismissed with professional reasons.

WriteUp has been built around that use case by a practising MRICS Chartered Surveyor. It reads a draft as one document, checking it against Red Book standards, instructions and configured criteria while leaving every judgement and decision with the valuer.

The useful test is straightforward: does the audit help the reviewer spend less time hunting for mismatched figures and more time considering whether the valuation rationale stands up under scrutiny? When it does, rules and AI are not competing methods. They are two forms of second pair of eyes, each applied where it is strongest.