Keller Williams needed a more accurate model to predict the likelihood that listings in various pipeline stages will eventually close.
We built a data aggregation pipeline that brings in different sources of data into a cohesive view of metadata and time series events for sales pipelines. We developed three types of models, one tree based model, and two types neural network models (one being embeddings + multilayer perceptron, and the other being embeddings + LSTM).
Model baseline was 15% more accurate than the existing statistical approaches.
Keller Williams needed a more accurate model to predict the likelihood that listings in various pipeline stages will eventually close.
We built a data aggregation pipeline that brings in different sources of data into a cohesive view of metadata and time series events for sales pipelines. We developed three types of models, one tree based model, and two types neural network models (one being embeddings + multilayer perceptron, and the other being embeddings + LSTM).
Model baseline was 15% more accurate than the existing statistical approaches.
