GSPANN’s Advanced Analytics Team created the ML algorithm utilizing its ‘Zeolite’ solution to predict when a customer is more likely to rebuy the same beauty product. For this purpose, we trained the ML model with historical data of customers. Before preparing the model, we performed multiple subsidiary tasks, including data extraction from various data sources, data exploration and research, data cleaning and wrangling, model building, and deploying the model in the client production system.
We compared model performance with the previous scheme by running A/B tests for six months. After ensuring performance superiority and consistency, we deployed it in production. Our Zeolite solution extracts three dimensions of data – customer data, product data, and transaction data stored in the client’s databases. We used Python scripts to clean data and create features required for model training using the “Gradient Boosted Machines” algorithm and predictions for everyday transactions.
The solution utilizes scheduled Python scripts for extracting data on a daily basis with the Control-M scheduler. Next, we transferred data to the saved model for predictions. Finally, we stored forecasts from the model in a new table of the database.