Human Judgement Versus AI Flags in Valuation

A valuation report can be technically sound and still contain one figure that does not travel through the document correctly. A £2,450,000 market value in the executive summary may appear as £2,500,000 in the security section. The conclusion may be right, but the discrepancy is still likely to prompt a lender query. That is where the discussion around human judgement versus AI flags needs to start: not with whether technology can value property, but with how it can help a valuer review the report before it leaves their desk.

A good review has always involved professional scepticism. The valuer considers whether the evidence supports the conclusion, whether the assumptions are reasonable and whether the report answers the instruction. Software does not replace that work. It can, however, read the report as one document, checking repeated figures, dates, terms and statements at a speed that is difficult to replicate manually when deadlines are tight.

Human judgement versus AI flags: different jobs

Professional judgement is required wherever a report calls for interpretation. Selecting comparable evidence, adjusting for location and condition, forming a view on market rent, applying a yield, and deciding the weight to place on a transaction are all matters for the registered valuer. A system can identify that one comparable reflects a 4.75% yield while the schedule states 5.25%. It cannot decide, without the valuer's reasoning, whether the lower yield is justified by covenant strength, lease length or an unusual lot size.

The distinction matters because a flag is not a finding of fault. It is a prompt to review. Some flags will be correct. Others will reflect a deliberate and properly explained departure from a pattern in the report. The value lies in ensuring that the point has been considered, not in forcing every figure into a uniform format.

This is particularly useful where the report has been amended over several drafts. A revised rent, floor area or lease expiry can be updated in the main valuation section but remain unchanged in a table, comparable commentary or appendix. The surveyor may know the correct position. The report still needs to state it consistently.

What an automated report audit can check well

An automated review is strongest when it deals with repeatable checks across a complete report. These are the areas most exposed to version changes, copy-and-paste amendments and simple pressure of time.

Figures that do not reconcile

Consider a commercial investment valued at £3,200,000. The report states a passing rent of £176,000 per annum and a 5.50% initial yield. Those figures broadly reconcile before costs, subject to the usual treatment of purchaser's costs and assumptions. If a later table gives the initial yield as 5.05%, the report may need an explanation or correction.

The same applies to areas. A valuation based on 1,850 sq m should not describe the subject as 18,500 sq m in the property summary. A zero added in error is obvious once seen, but it can be missed when the author is reviewing familiar text. A report audit can compare repeated numerical references and flag material differences for the valuer to assess.

It can also check calculations that are stated in the report. If market rent is expressed as £215 per sq m on an area of 940 sq m, the implied annual rent is £202,100. Where the report instead states £220,000, the valuer can quickly establish whether the area, rate or annual figure is the intended input.

Dates, lease terms and tenant statements

Lease information often appears in several places: the tenancy schedule, valuation rationale, comparable analysis and security comments. A lease expiry of 24 June 2031 should not become 2032 in the narrative unless there is a separate reversionary or break assumption being discussed.

The same principle applies to unexpired terms. If the report was valued as at 31 March 2026 and the lease expires in March 2036, a statement of "approximately 12 years unexpired" needs checking. It may be a harmless legacy sentence from an earlier draft, but it affects the reader's understanding of the investment.

Tenant covenant commentary is a further example. A report may correctly identify the named tenant in the schedule, then refer to a previous occupier in a paragraph copied from historic reporting. The valuer remains responsible for the covenant assessment. A flag simply draws attention to a contradiction that could otherwise distract from the analysis.

Instruction-specific requirements

Red Book compliance and lender instructions are not interchangeable checklists. The requirements depend on the instruction, property and purpose of the valuation. That is precisely why the valuer must remain in control.

A structured audit can nevertheless check whether defined requirements appear to have been addressed. For a residential block, this may include whether an EWS1 disclosure required by the instruction has been considered. For a secured lending report, it may include whether the stated marketing period is consistent with the instruction, or whether the required minimum comparable evidence is present.

A missing disclosure is not proof that the underlying issue has been ignored. It may be addressed in supporting papers or be genuinely inapplicable. But if the instruction calls for a statement in the report, a flag gives the author the chance to make the position clear before submission.

Why manual review alone has limits

Manual review remains essential, especially for judgement, tone and the relationship between evidence and conclusion. Yet it has predictable limits. Reading your own work encourages confirmation bias. You tend to see the sentence you intended to write, particularly after several rounds of amendment.

A colleague's review can provide useful challenge, but it takes time and is not always proportionate for every instruction. Smaller teams also face the practical issue of independence: the person most able to spot a valuation issue may be the person who prepared the report.

Technology is not free from limitations either. It needs to be configured against the relevant reporting requirements, and its output needs to be read critically. A flag without context can create unnecessary work. An overly broad rule can confuse a legitimate judgement call with an inconsistency. The aim is not to produce a long list of warnings. It is to produce a focused review of points that deserve a second look.

A practical way to use AI flags before submission

The most effective process is to use an automated audit after the draft is substantively complete, not as a substitute for drafting or valuation analysis. The valuer prepares the report, reaches the valuation conclusion and checks the core reasoning in the usual way. The report is then reviewed as a whole for contradictions, calculations and instruction-specific omissions.

Each flag should be dealt with deliberately. Confirm the point, amend the report where necessary, or record why the wording or figure is correct. For example, a different market rent in two sections may be intentional where one refers to an assumed rental value and the other to a current headline rent. The report should make that distinction explicit rather than merely clearing the flag.

This creates a useful audit trail of professional consideration. It does not change who makes the decision. The registered valuer decides whether the evidence supports the conclusion and whether the final report is suitable for issue.

Data control and professional responsibility

For valuation practices, the question is not simply whether a tool can identify errors. It is what happens to the report data while it does so. Client confidentiality, security and control need to be considered before any report is uploaded for review.

The sensible standard is private, encrypted processing, with report data not retained or used to train public models. The provider should be clear about where data is processed, who can access it and how long it is held. Those questions are part of normal supplier due diligence, not an obstacle to using technology.

WriteUp was built by a practising MRICS Chartered Surveyor and operates as a professionally informed second pair of eyes for valuation reports. Its role is to flag figures that do not reconcile, missing requirements and report-wide contradictions. The surveyor retains control of every finding and every final decision.

The strongest reports are not those with the fewest flags. They are the reports in which each material point has been considered, the evidence is clearly connected to the conclusion, and the final wording stands up when someone else reads it closely.