Propensity Modeling
Project Overview
Identifying high-potential customers is crucial for strategic marketing and promotional events to maximize return on investment. This project aims to evaluate customer spending power using information from an existing database.
Challenge
The primary objective was to train a predictive model to estimate the spending scores of the customers in an existing database. In the training dataset, spending scores were manually assigned by the business owner based on customer behavior and spending patterns. A higher score indicated greater purchasing power and propensity.
Approach
- Algorithm Design: Develop a prototype neural network model for propensity modeling.
- Data Extraction: Pipeline the required data from the customer database into Python.
- Data Pre-processing: Clean and split the data into training, validation and testing datasets.
- Training and Iteration: Train a neural network model on the training dataset, and iterate over previous steps to improve performance.
- Optimization: Select the best-performing model by fine-tuning the hyperparameters.
- Evaluation: Report the performance of the model on the testing dataset.
- Deployment: Implement the propensity model using cloud services, connecting the model to the customer database and other follow-up applications.
Tools
The project used Pandas for data processing and PyTorch for training the neural network model in Python. Git and GitHub were employed for version control and collaboration. Testing and deployment were facilitated by continuous integration/continuous deployment (CI/CD) practices. The Google Cloud Platform (GCP) was utilized for cloud deployment.
Results
The final product was a fully trained neural network model that predicted propensity score for existing customers in the database. The target audience was accurately identified for follow-up actions. This resulted in enhanced customer relationship management and improved outcomes for marketing events.