Case Study: Welsh Government
Welsh Government Land Valuation Research
Published . Findings directly inform Welsh Government local government finance policy and are cited in a Welsh Government Ministerial Written Statement.
Coverage
1,912
Lower Super Output Areas
Geography
99%
of Welsh LSOAs covered
Methodologies Tested
5
distinct valuation approaches
The Challenge
Welsh Government needed to evaluate different approaches for assessing land values to support local government finance policy, specifically to inform potential reform of the council tax and non-domestic rates system in Wales. Existing methods lacked a systematic comparison across the full Welsh geography. The project was commissioned under Welsh Government procurement standards and published on GOV.WALES.
The core question: could AI-driven methodologies outperform traditional approaches, and which method would best support equitable, evidence-based finance policy? The work aligned with priorities set by the Office for National Statistics (ONS) on subnational economic statistics and the Office for Statistics Regulation (OSR) framework for assured public statistics. The Department for Science Innovation and Technology (DSIT) national AI strategy also identifies geospatial and property data as a priority AI application domain.
Five Methodologies Tested
The analysis spans 1,912 Lower Super Output Areas (LSOAs) with land transactions, the small-area geographic unit defined and maintained by the Office for National Statistics, which is around 99% of all Welsh LSOAs, enabling direct, like-for-like comparison consistent with UK-wide subnational data standards.
| # | Methodology | Approach |
|---|---|---|
| 1 | Market-Based Statistical Valuation | Regression and hedonic pricing models anchored to observed market transactions. |
| 2 | Advanced Algorithmic and Machine-Learning Applications | Gradient boosting, random forests, and neural network models trained on land-use and transaction data. |
| 3 | Formula-Based Valuation by Land Area | Rule-based per-hectare and per-unit formulae applied uniformly across geography. |
| 4 | Conventional Valuation Approaches | Professional surveyor methodologies consistent with RICS Red Book standards. |
| 5 | Innovative Experimental Approaches | Novel hybrid methods combining spatial econometrics with satellite-derived land-cover data. |
Assurance
- •Delivered under Welsh Government procurement standards.
- •Full summary and comprehensive report published on GOV.WALES.
- •Cited in a Welsh Government Ministerial Written Statement (17 March 2026).
Outcome
Published March 2026. The comparative analysis across all five methodologies directly informs Welsh Government local government finance policy. Of the five methodologies tested, the machine-learning method (Gradient Boosting / CatBoost) delivered the best overall predictive performance, though the report is candid that all five remain below the accuracy required for statutory valuation. Findings align with HM Treasury Green Book guidance on appraisal and evaluation.
Findings are cited in a Welsh Government Ministerial Written Statement (17 March 2026). The full report is publicly available on GOV.WALES and as a downloadable PDF.
"The Welsh Government land valuation research demonstrates exactly what mixed-methods AI analysis can achieve. Combining machine learning models with conventional valuation techniques across 1,912 lower super output areas gave Welsh Government a genuinely comparative evidence base for local finance policy reform."
Dr Stylianos Kampakis, Managing Director, Tesseract Academy
Access the full report
Published on GOV.WALES. Also available as a PDF download.
