With just four months to deliver 15 predictive models for key go-to-market (GTM) plays of specific product offering groupings, Revel assembled a nimble yet capable team of data scientists, technical architects, and others, and got right to work.
The Revel team used more than 1,500 features (or data attributes) from six groupings of data sources. As no historical data existed to perfectly represent the key GTM plays, many features were created by proxy, greatly increasing the complexity. To solve for this, the team leveraged Microsoft SSAS to create tabular data models and Jupyter Notebook for feature engineering.
After defining the data features of each GTM play, the team used XG Boost and Auto-Sklearn in Python to develop the machine learning models and auto_ml to accelerate hyperparameter tuning to make up for time invested in feature engineering. The entire solution was developed within Docker and Windows Subsystem for Linux (WSL). With accurate predictions created, the team employed Microsoft Power BI to provide a visualization of model results and deliver meaningful insights to the Service group’s leadership team.