On this page
Do AI Valuation Models Remove the Need for a Site Visit?
A Zimbabwean bank approving a mortgage, a pension fund revaluing its portfolio, and an executor settling a deceased estate all need the same thing: a value they can defend. Registered valuers produce that value daily, but in our experience the evidence behind each report stays in a client file, with no shared dataset others can draw on. Automated valuation models (AVMs) promise to change that by estimating value from data in seconds.
The answer to the question valuers are asking is no. An AI valuation model does not remove the need for a site visit, and the International Valuation Standards Council (IVSC) states that no model can produce an IVS-compliant valuation without the valuer's professional judgement. The standards do not ban desktop work outright, though. They allow agreed limits on the valuer's investigations, provided those limits are noted in the report. That is the space where AVM-assisted valuation can legitimately operate.
This article covers what AVMs need to work, what only a site visit can see, where Zimbabwe stands, why the industry must collaborate on data, and how valuers can use AI tools without lowering the standard of their reports.
How Property Valuation Works in Zimbabwe Today
Who Can Value Property and Under Which Standards
The Valuers Act [Chapter 27:18] establishes the Valuers Council and provides for the registration of valuers and the regulation of their practice. The Valuers Council requires IVS compliance for professional practice, particularly in property and asset valuation, as the Zimbabwe Independent has reported.
Any AI-assisted valuation produced in Zimbabwe is therefore judged against IVS, not against a software vendor's marketing claims.
Why Zimbabwean Valuation Evidence Does Not Accumulate
Every valuation starts with comparable evidence, and in Zimbabwe that evidence is thin. A Zimbabwe Independent commentary notes that scarce comparable transactions make accurate valuations harder to obtain. The Deeds Office records the transfer of a property, but that data is not digitised in a format that feeds market analytics, and no portal, government body or industry association maintains a usable transaction price history.
Sparkline's earlier analysis concluded that this leaves no accessible, searchable property sale price database from which a valuation model could be trained. In our experience, valuers largely rebuild the evidence base on each instruction.
What Automated Valuation Models Are and How They Estimate Property Value
Comparables-Based and Hedonic AVMs Work Differently
An AVM is a statistical system that estimates a property's value at a given date without a person performing the analysis. Comparables-based models select similar sales for each subject property, so their results can be traced, while hedonic models insert property characteristics into fixed equations and cannot be traced the same way. Machine-learning models marketed as AI valuation tools sit on top of the same data requirements.
A valuer who must defend a number to a lender or a court needs to see what produced it.
How Accurate AVMs Are and Why Confidence Scores Matter
The standard accuracy measure is the median absolute percentage error. In US lender data cited by AmeriSave, well-performing AVMs with high confidence scores reach median errors of roughly 3% to 7%, while lower-confidence properties can see errors widen past 15% or even 20%. Those figures come from a data-rich market, so they describe a ceiling for Zimbabwe, not a forecast.
South Africa's Lightstone publishes two scores with each valuation. The accuracy score expresses confidence that the estimate falls within 20% of the likely selling price, and the safety score expresses confidence that the model is not over-predicting by more than 10%. Lightstone builds on deeds records, cadastral data, municipal rolls and listing data, and describes its design as human-in-the-loop, with the confidence score signalling when human review is advisable. A model that states its own uncertainty gives a valuer a rational rule for deciding which properties need a visit.
Why Site Visits Remain Essential in AI-Assisted Property Valuation
What IVS and RICS Require When a Model Is Used
IVS 105, a standard on valuation models, was added in the edition effective 31 January 2025 and stipulates that models must be supplemented by professional judgement to achieve IVS compliance. RICS wrote the same principle into its Red Book: outputs from AI, AVMs or valuation software count as a written valuation only if the valuer has applied professional judgement to them.
Limited-inspection work remains possible under these rules. IVS has long accepted agreed limits on investigation, but limits so substantial that the valuer cannot assemble sufficient evidence mean the valuation does not comply with IVS. A desktop or AVM-assisted valuation is therefore acceptable when the client agrees, the limits are stated, and the evidence is still adequate for the purpose. An unreviewed model output meets none of those tests.
Four Things an AVM Cannot See From a Desk
Academic work published by TEGoVA lists limited ability to address a property's condition and limited ability to account for external influences among the well-known limitations of AVMs, and notes that models implicitly assume marketable condition. In Zimbabwe, these gaps coincide with the areas where local data is weakest.
Factor | What the model works from | What only a site visit confirms | Zimbabwean example |
|---|---|---|---|
Hidden physical condition | An assumption of typical, marketable condition | Structural defects, dampness, deferred maintenance, quality of internal finishes | A house with fresh paint over damp walls and an ageing roof |
Unique property features | Standardised fields from deeds, listings and public data | Custom renovations, architectural features, plot configuration, installed utilities | A property with a high-yield borehole and full solar installation next to one with neither |
Unrecorded local factors | Maps and databases | Noisy adjacent commercial use, traffic, ongoing construction, access road condition | A residential stand next to a house now operating as commercial premises |
Professional verification | Nothing: the output is not yet a valuation | The valuer's inspection, judgement and sign-off required by IVS and the Red Book | An extension built without council approval that no record shows |
Local listing data widens the gap. Our earlier analysis found that Zimbabwean listings frequently lack standardised fields for water source, electricity connection and solar presence. A model has nothing to read for exactly the features that move value locally.
How AI Valuation Models Help Zimbabwean Valuers
AI Tools Move Valuer Time From Data Collection to Judgement
The IVSC presents AI as support for the valuer: by freeing up time, these tools let valuers concentrate on analysis, scepticism and judgement. The repetitive work is concrete: searching past instructions for comparables, re-keying property particulars into reports, and checking reports for internal consistency. A confidence score then lets a practice sort its workload, sending standard, high-confidence properties down a documented desktop route and unusual or low-confidence ones straight to full inspection.
Model-Assisted Monitoring Keeps Lenders' Risk Process Inside the Profession
Zimbabwe's banking sector remains substantially collateral-based, per a Zimbabwe Independent commentary. A bank with hundreds of mortgages cannot send a valuer to every property each year, so between-valuation monitoring becomes necessary. If valuers supply it as an AVM-assisted desktop service with sign-off and stated limits, the bank's risk process stays within the profession.
How Valuers Can Meet IVS and RICS Requirements When Using AI Tools
When to Use Desktop, Model-Assisted or Full Inspection Routes
The matrix below is Sparkline Labs’ suggested starting framework, not a Valuers Council rule.
Situation | Suggested route | Reason |
|---|---|---|
Monitoring performing mortgages on standard houses with a high confidence score | Model-assisted desktop review with valuer sign-off and stated limits | The purpose is monitoring, and condition is unlikely to have changed materially |
New lending, purchase decisions, disputes or litigation | Full inspection | The evidence must withstand challenge |
Unusual, commercial, industrial or farm property | Full inspection, with the model used only as a cross-check | Few comparables exist, so model output is unreliable |
Low confidence score or sparse data in the area | Full inspection regardless of purpose | The model itself signals insufficient evidence |
Five Controls That Keep AI-Assisted Valuations Defensible
- State in the scope of work and the report where a model was used, what it produced and what was not inspected, since the IVSC expects future standards to require greater transparency about where and how AI is used.
- Test the model before relying on it, because IVS 105 requires the valuer to apply professional judgement in balancing model characteristics such as accuracy and completeness.
- Check model inputs against your own knowledge of the property and suburb, since valuation models are only as strong as the information they learn from.
- Record every point where you overrode or adjusted the model output and the reason, because that record is the evidence of the professional judgement the Red Book requires.
- Keep a structured, confidential record of each valuation's inputs and conclusions so the practice builds its own comparables dataset instead of re-keying evidence each time.
Valuers with RICS credentials should also check the status of new guidance. RICS planned to issue global practice guidance on AI in real estate valuation for consultation in Q2 2026, with publication expected later in 2026.
Where Zimbabwe Stands on AI Valuation and Valuation Data
Zimbabwe Has No Public Property Price Index or Documented AVM Yet
As far as we could establish, no automated valuation model is documented as operating in Zimbabwe, and we found no public property price index. The comparison with South Africa shows the gap.
Requirement | South Africa | Zimbabwe |
|---|---|---|
Transaction and title data | Deeds records, cadastral data and municipal rolls combined in one model | Deeds Office transfers are not digitised in a form that feeds analytics, and no usable sale price history exists |
Independent standards | Lightstone is a member of the European AVM Alliance | No equivalent local model standard identified |
Confidence scoring | Accuracy and safety scores attached to each valuation | No model in public use to attach them to |
Shared property index | Built on national data coverage | None identified |
Deeds Digitisation and the National AI Strategy Are the Nearest Policy Levers
Zimbabwe's National AI Strategy 2026 to 2030 lists real estate among additional sectors and expects it to build its own data layer through private initiative. The Deeds Registries (General) Regulations 2025 introduce deed validation and digitisation with a July 2027 compliance deadline, which could improve transaction records but depends on seller cooperation. Neither creates a shared valuation dataset on its own.
Why Zimbabwe's Property Industry Needs to Collaborate on Data
Valuers cannot close the data gap on their own, and neither can any other single party. Agents, banks, valuers, councils and portals each hold a fragment of what a valuation model needs, and the fragments do not fit together. Collaboration here means agreeing common field definitions, contributing anonymised outcomes, and giving the result to a neutral custodian.
Why No Single Valuer, Agency or Portal Can Build a Valuation Dataset Alone
One firm's records cover its own instructions, which skew toward its own clients, suburbs and property types. A model trained on that sample learns the firm's book, not the market. Our earlier analysis identified three separate fragmentation problems: portals use incompatible field structures, agency registration is not tied to any data quality requirement, and transaction prices are not published anywhere accessible. Fixing one without the others leaves the dataset unusable.
How Kenya Built a Property Index From Pooled Data
Kenya shows the sequence. The private Hass Index drew on Hass's own data, more than 30 other Nairobi estate agencies, online portals and press listings to build a composite price series, and it is based on asking prices rather than completed sales. In 2026 the Kenya National Bureau of Statistics published its first official, quality-adjusted residential property price index, using listings, a real estate survey and quarterly surveys of agents, with a hedonic method that controls for differences between properties.
Pooled private data came first and an official index followed. Valuers hold completed-valuation evidence that asking-price indices lack.
What a Shared Data Standard Does: The RESO Example
In the United States, the Real Estate Standards Organization (RESO) maintains the Data Dictionary, which defines field names, data types and allowable values for listing data across MLS systems, and it certifies systems for compliance. Zimbabwe needs nothing on that scale. It needs a minimum shared vocabulary for suburb, floor area, land size, water source, electricity, solar, construction and condition, agreed once and used everywhere.
What Collaboration Looks Like in Practice
Four steps would move the industry from fragments to a usable dataset.
- Agree a minimum field set for valuations and listings, held by a neutral body such as a professional institute or regulator, so every participant records the same attributes the same way.
- Contribute anonymised valuation outcomes, with date, suburb, property attributes, value and basis of value, subject to client confidentiality and data protection law.
- Link valuation records to digitised Deeds Office transfers as that process matures, so asking prices and valuations can be checked against completed sales.
- Publish the methodology and confidence measures so lenders, agents and valuers can judge how far to trust any index or model built on the data.
Why Shared Property Data Matters Beyond Valuation
A shared dataset serves everyone who touches a transaction. Valuers gain comparables they do not have to rebuild, lenders gain better collateral monitoring, agents price against evidence instead of anecdote, and buyers abroad gain a reference point that does not depend on the seller's account.
Why Zimbabwean Valuers Should Care About AI Valuation Models Now
Valuers Hold the Missing Training Data
The scarcest input in Zimbabwe's property market is not software. It is structured evidence of what properties are worth. Valuers produce it continuously, which makes them the natural contributors to a shared dataset.
A Signed Report Carries the Valuer's Liability, Not the Model's
Gadsden Valuations notes that in UK practice the Registered Valuer who uses an AVM output within a written valuation retains professional responsibility. IVS and the Red Book point the same way, because a model cannot sign a report or answer a challenge. A valuer who relies on an unvalidated tool inherits its errors under their own registration.
What a Valuation Practice Built for AI Looks Like
The site visit is the most information-rich event in a valuation, and most of what it produces is stored as prose. A practice designed for AI treats the inspection as structured data capture: condition, utilities, features, local influences and photographs recorded once, in fixed fields, and reused for the report, the model and the next comparables search.
That is an engineering problem before it is an AI problem. The same principle sits behind Propertyzone, where structured field completion is enforced at submission so incomplete records never enter the dataset. A valuation practice starting this work would begin with solution architecture, defining the fields and workflow before choosing any model. AI will not remove the site visit. It will decide whether the evidence gathered there is used once or many times.
Sources
- International Valuation Standards Council. (2025). Navigating the Rise of AI in Valuation: Opportunities, Risks, and Standards. IVSC.
- CBV Institute. International Valuation Standards: What They Are and What CBVs Should Know (earlier IVS edition).
- Government of Zimbabwe. Valuers Act [Chapter 27:18]. ZimLII.
- The Zimbabwe Independent. (2026). Accountants, valuers: Twin guardians of financial truth.
- The Zimbabwe Independent. (2024). Perspectives: Navigating the real estate valuation process in Zimbabwe.
- Sparkline Labs. (2026). AI Readiness in Zimbabwe Real Estate 2026: Why Your Listing Data Is Blocking Every Application That Matters.
- Wikipedia. Automated valuation model.
- AmeriSave. (2026). Automated Valuation Models in 2026: How AVMs Work, Their Limits, and What They Mean for Your Mortgage.
- Cover. Quick and easy asset valuation (describing the Lightstone AVM).
- Lightstone Property. (2026, April 28). AI-driven valuations transform SA's property market.
- DealerFloor. Lightstone's approach to transparent, context aware AI.
- International Valuation Standards Council. (2024, January). International Valuation Standards effective 31 January 2025 (hosted copy). Fondazione OIV.
- Property Elite. (2025). RICS Valuation: Global Standards (Red Book, January 2025) summary.
- Matysiak, G. A. Assessing the accuracy of individual property values estimated by automated valuation models (title truncated in source). TEGoVA.
- AVAA. (2025, September 3). IVSC Artificial Intelligence In Valuations Paper.
- The Zimbabwe Independent. (2026). Treating valuation as invisible infrastructure of our economy.
- Royal Institution of Chartered Surveyors. AI in real estate valuation (global practice guidance, forthcoming).
- Gadsden Valuations. What Are AVMs?
- HassConsult. About the Hass Index.
- Kenya National Bureau of Statistics. (2026). Kenya Residential Property Price Index, April 2026.
- Kenya Mortgage Refinance Company. (2026). Research report Q1 2026 (methodological note on the KNBS index).
- Real Estate Standards Organization. RESO Data Dictionary FAQ.
- Real Estate Standards Organization. Certification and MLS Map.
About the Author
Building Something Like This?
We build platforms, internal tools, and integrations for Zimbabwean and African businesses. Bring us the problem, the idea, or the system that is no longer working as it should.