Research · Thread 07
Data Centre
An on-demand automated valuation model reconciling disagreeing Dubai real-estate data sources into one auditable number.
Valuation in Dubai real estate runs on analyst cycles: pull comps, cross-reference several registries, hand-check the number. Data Centre asks whether an on-demand AVM can replace that with an auditable number in seconds, and the hard part isn't the model. It's reconciling the 9 disagreeing sources that completed an ingest, of 115 connectors registered, into one confidence-tiered entity ontology across the deployed snapshot's 40K+ entities before any valuation math runs.
Finding
An auditable number in seconds
The question is whether an on-demand valuation can replace that cycle with a number produced in seconds that still shows its working, so a broker can see which sources it leaned on rather than being asked to trust it.
The hard part isn't the model, it's reconciling the sources before any valuation math runs.
Paper to shipped
- Confidence-tiered entity ontology, 40K+ entities in the deployed snapshot
Sources disagree on the same entity constantly. Resolving them into a tier-scored ontology (T1 government registry, T2 developer and official, T3 computed and market, T4 broker and editorial) makes the disagreement visible instead of averaging it away.
- MAPE/RMSE calibration and backtest harness
An internal snapshot showed a mean absolute percentage error of 24.8% (leave-one-out backtest, July 2026) over 121 held-out sales, published in the tool's own UI rather than a hidden report.
- Data Centre, the valuation surface
React + MapLibre valuation surface the team opened daily: on-demand AVM brief, ranked opportunities, six-month community-calibrated forecasts, and a risk-adjusted investment score that separates the property-specific signals from the fixed macro posture.
