PropTech UK Residential AVM SaaS -- RICS and FCA MCOB

Data sources

HMLR Price Paid Data (30M+ UK residential transactions), Land Registry Title data, EPC Register (MEES -- significant value driver), OS AddressBase Premium (UPRN geospatial), ONS Census data (ACORN neighbourhood), Zoopla listing data (current market).

Model

• XGBoost gradient boosting -- outperforms hedonic regression at scale.
• Every AVM output includes 80% and 95% confidence intervals -- mandatory per RICS guidance.

RICS AVM Guidance Note Compliance

RICS AVM guidance (January 2023): (1) transparency (methodology documented and explainable -- SHAP values showing which features drove each valuation), (2) confidence scoring (displayed prominently), (3) comparables disclosure (5 most similar transactions available on request), (4) limitations disclosure (AVM is estimate, not RICS Red Book valuation -- clearly stated), (5) data freshness (regular retraining -- stale training data = unreliable AVM).

FCA MCOB for mortgage lenders

• 1AVM for remortgage (LTV below 75% -- AVM acceptable),
• 2AVM for additional borrowing (existing lender),
• 3desktop valuation (mid-range LTV),
• 4full RICS Red Book (high LTV, large loans -- AVM insufficient alone).

AVM API response

• value + confidence score + confidence interval + 5 comparables + RICS compliance confirmation.
• Quarterly model performance report for lender MCOB compliance file.

Weekly comparison

• AVM predictions vs actual HMLR sold prices.
• MAPE tracking (target: below 5%).
• Geographic MAPE by region.
• Automated retraining on SageMaker when MAPE exceeds threshold.
• Champion-challenger: new model vs production -- if challenger wins 4 consecutive weeks, promote.

COVID drift risk

models trained 2019-2022 may underperform in 2025-2026 market -- continuous monitoring essential.