One of our American real estate clients faced the problem of low sales. That’s why they decided to boost the number of estate buyers via ML-driven targeted advertising. We used historical sales data on transactions, loans, and estimated property value, to build an ML model for highly targeted advertising. See how we beat challenges like this.
First, we used ATTOM datasets (the US nationwide property data) related to the ownership status and seasonality. So we could move on to create an optimal prediction model, considering sales fluctuations. Taking the time of ownership into account helped in targeted advertising. As a result, we got a five times growth in estate sales vs. random advertising (1422 vs. 281).
Adding the owner's actual residence data also improves the targeting of the ads sent. So it let us create a Facebook-like targeting model, but a more precise one, considering specific data related. Consequently, we reached a 38 times increase in house sales (1064 vs. 28).
We choose the following parameters for the model: the period of ownership, the equity position, and actual residence became the ground for the prediction model. Later we ensured that our model was robust enough to make further predictions, avoiding the "black box" effect. It has the decision tree classifier under the hood to make decisions traceable.
Frameworks and Libraries
AI Development Stack
We improved ad targeting enabling us to increase sales conversion 16,5 times, so our client was happy. The figures speak for themselves: 2408 houses sold using the prediction model vs. 146 when randomly sending ads.
So this is how we created a marketing forecasting solution for real estate businesses increasing house sales 16,5 times per month.
