Machine-trained valuation in real estate
Context & Objectives
A leading real estate services company was missing out on high-margin opportunities in its commercial efforts. It was still relying mostly on human « gut feeling » to appraise for sale and for rent properties.
Its existing tool was logging deal values but offered little in the way of recommendation or valuation. Furthermore, data quality was a persistent issue.
We thus set out to design a customized valuation system.
Approach
Cleaning and homogenizing the property and deal databases laid the project’s groundwork.
From there on, we integrated business knowledge into the tool through feature engineering and incorporated real estate experts’ pricing criteria.
We then compared two solutions to estimate the technical feasibility/efficiency of the property appraisal
The first one used a simple regression model, allowing the client to understand the outcome’s calculations fully.
In the second, a machine-learning component evaluated the house’s properties in combination with expert-defined criteria. While more complex and thus less “explainable” to the untrained eye, it provided more accurate estimations.
Our client opted for the second solution in the end.
Results
In less than four weeks, we provided our client with an automated tool ready for deployment, whose property appraisal accuracy rate was good enough to provide a significant competitive advantage against the competition in terms of:
Data-driven quality of appraisal
Speed of first appraisal
